feat(web): enhanced file panel — side-by-side diff, hide whitespace, inline review

Adds DiffSplitView component for side-by-side diff mode, whitespace-only
change filtering, inline review comments with thread/gutter cell UI, diff
preferences persistence, and write-file API support for in-browser editing.

Backend: hideWhitespace param on git diff endpoint, write_file route.

.gitignore

@@ -1,6 +1,8 @@
 node_modules
 dist
 .env
+.env.*
+!.env.example
 
 # Claude / Cursor (local agent & IDE config — CLAUDE.md and AGENTS.md stay tracked)
 .claude/
@@ -18,3 +20,4 @@ data/*
 !data/mcp.example.json
 !data/coder-providers.example.json
 codecontext/fork.tar.gz
+/Arena

CHANGELOG.md

@@ -2,6 +2,36 @@
 
 All notable changes per release tag. Most recent on top, ordered by tag creation date (which matches the git history). Tag names follow `vMAJOR.MINOR.PATCH-slug` — the slug describes what shipped, so the tag name alone is enough to recall the batch.
 
+## v2.8.0-fork-lifts — 2026-06-07
+
+Completes the eight fork-lift integrations from `/opt/forks` into BooCode: boocontext sidecar upgrade, LSP code intelligence, DCP clean-room pruning, institutional memory, subagent protocol enhancements, plugin hook host, inference reliability (tool-shim + loop detectors), and TokenScope token breakdown. Backfills edit safety guards (truncation + dropped imports) and the TokenScope analyzer/persist module. Closes the fork-lifts-mit epic.
+
+**boocontext sidecar (Phase 3):** Upgrades the `codecontext` container from the old Go MCP server to the boocontext Node.js MCP aggregator. Multi-stage Dockerfile builds boocontext from `/opt/forks/boocontext` alongside the HTTP shim. `shim.go` gains `CODECONTEXT_CHILD` env-var support and three new HTTP routes for symbols, callgraph, and blast radius. Three TypeScript tool wrappers (`get_symbol_details`, `get_call_graph`, `get_blast_radius`) registered on the server, with blast radius added to the synthesis pipeline. Docker-compose env vars configure child MCP paths (tree-sitter-analyzer, type-inject).
+
+**LSP integration (Phase 4):** Six-file `lsp/` module in the coder with config, JSON-RPC stdio client, lazy server-manager (per-project pool, 5-min idle shutdown), and operations (diagnostics, goto-definition, find-references). Three read-only agent tools registered — `lsp_diagnostics`, `lsp_goto_definition`, `lsp_find_references`. TypeScript/JavaScript only in v1.
+
+**DCP clean-room (Phase 5):** Seven-file `dcp/` module in the server inference pipeline. Consecutive identical tool_call+tool_result pairs are deduplicated; failed/empty tool results are purged via configurable window. Orchestrated by `transformMessages()` running before `buildMessagesPayload` in `turn.ts`. Clean-room reimplementation — AGPL source was referenced for behavior only. 10 unit tests.
+
+**Institutional memory (Phase 6):** Eight-file `memory/` module with file-based recall. Hierarchical 4-scope scan (global → home → project → session) under `.boocode/memory/`. Keyword/tag relevance matching at prompt assembly. Injected as a `<boocode-memory>` block in the system prompt. v1 recall-only — extract/dream deferred.
+
+**Subagent protocol (Phase 7):** `AgentCapabilitiesSchema` in contracts with `supportsStreaming`, `supportsReasoningStream`, `supportsBackgroundExecution` flags. `ProviderSnapshotEntry` gains the two streaming capability fields. `new_task` tool gets a `background` mode flag for non-blocking dispatch. Flow-runner already supported per-step model override.
+
+**Plugin host (Phase 8):** Typed hook registry in `plugins/host.ts` with `registerHook`/`emitHook` for five lifecycle events: `tool.execute.before`, `tool.execute.after`, `turn.start`, `turn.end`, `task.terminal`. Patterns-only from oh-my-openagent (SUL — no code copy).
+
+**Inference reliability (Phase 9):** `tool-shim.ts` recovers XML/JSON tool calls from plain-text model output (e.g. Qwen inline format). `loop-detectors.ts` catches content-repeat and tool-loop patterns. Existing doom-loop detection remains — detectors are additive.
+
+**Edit safety guards (Wave 1):** `edit-guards.ts` rejects catastrophic truncation (>60% chars AND >50% lines). `edit-guards-imports.ts` detects dropped import statements. Both run in `pending_changes.ts` immediately before `writeFileAtomic`.
+
+**TokenScope (Wave 2):** `TokenBreakdownSchema` in contracts with system/user/assistant/tools/reasoning categories. `token-analysis/` module with analyzer and DB persistence. `ContestantShape.token_breakdown` field and `token_breakdown` JSONB column on `contestants`/`tasks` tables. Arena `computeBenchmark` accepts and returns token breakdown.
+
+**Build:** Server 649 ✅ Coder 471 ✅ Contracts ✅ — all green.
+
+Adds the **Arena** pane for running the same prompt against 2–6 AI competitors simultaneously and picking the best result. A Battle is one Arena run: pick a battle type (Coding — backend+model with git worktrees producing diffs; or Q&A — BooChat persona+model producing text), write or generate a prompt, add contestants, and hit Start. Contestants are scheduled in two concurrent lanes — the local lane (llama-swap models, serial) and the cloud lane (Claude Code, OpenCode-on-cloud, parallel). The lane scheduler captures wall-clock duration for every contestant and tokens/sec for local models. When all contestants finish, a two-stage analysis (digest then judge) auto-runs on the DEFAULT_MODEL, writing `analysis.md` naming a winner; the user can override the winner per-row or trigger cross-examination. Results land in `/<project-root>/Arena/<dated-battle>/` with per-contestant `result.md`, diff patches for coding, and `manifest.json`. Replaces the old API-only `POST /api/arena` with dedicated `battles`/`contestants`/`cross_examinations` tables and full UI. Also adds a `DiffView` component with line-by-line colored unified diff and a per-row dropdown for winner override. Built on `v2.7.18-permission-modes`; pairs conceptually with the earlier `v2.7.17-orchestrator` multi-agent work (both share the pane kind pattern and `onTaskTerminal` hook).
+
+## v2.7.18-permission-modes — 2026-06-05
+
+Adds a unified **permission picker** to the BooCoder composer — Plan / Ask Permission / Bypass — replacing the old raw per-agent mode dropdown that exposed each agent's full native vocabulary with inconsistent labels. The three options map generically onto every provider's existing mode metadata: the `plan`-id mode → Plan, the default mode → Ask, the `isUnattended` mode → Bypass (claude `bypassPermissions`, qwen `yolo`, opencode `full-access`); goose has no modes so it shows no picker, exactly as before. `modeId` stays the single wire field — the active unified mode is derived from it, so no contracts change was needed. Native BooCode gains its own mode set (registered in the manifest and exposed by the snapshot): **Ask** stages edits to the pending-changes queue as today, **Bypass** auto-applies the queue to disk after the turn (both the interactive messages path and the task-based dispatcher path), and **Plan** falls back to Ask — the shared `apps/server` inference engine is deliberately left untouched. A supporting fix preserves the `isUnattended` flag on live-probed ACP modes (`acp-derive.ts`) so opencode's bypass mode is still detectable from the wire. Coder 373 tests green, coder + web typecheck clean. Built on `v2.7.17-orchestrator`.
+
 ## v2.7.17-orchestrator — 2026-06-03
 
 Brings the deterministic multi-agent "conductor" into the app as the **Orchestrator**: launch any read-only Han flow (research, code-review, investigate, architectural-analysis, security-review, …) from BooChat or BooCoder and watch each specialist agent stream live in a Paseo-style run pane, ending with an evidence-disciplined, adversarially-validated report — all on free local Qwen, persisted and resumable. Built and audited end-to-end via `paseo-epic` in an isolated worktree, on top of the prior `/opt/boocode/conductor` standalone CLI: the conductor's 22 flow definitions, Spine factory, and Han evidence/YAGNI contracts were re-homed into `apps/coder/src/conductor`, and a new DB-backed flow-runner (`flow_runs`/`flow_steps`) dispatches each step as a real BooCoder task through the existing dispatcher — reusing its streaming→WS-frame pipeline and worktree-as-read-snapshot, with an `onTaskTerminal` hook that advances the wave and a startup resume that re-dispatches in-flight steps after a coder restart. Read-only is enforced hard: every step is dispatched `qwen --approval-mode plan`, an adversarial-security review caught and closed a bypass where a qwen-unavailable task silently fell through to write-capable native inference (now fails closed), and the ACP path's mode-set was made fail-closed too. The UI adds a fourth `orchestrator` pane kind (collapsed agent roster, expand-one live stream, report on top), a Workflow button + slash flows on the shared `ChatInput` for full BooChat/BooCoder parity, a "New Orchestrator" entry in the + and split menus, a category-grouped launcher dialog, runs history, and export (copy / save-to-file / send-to-chat) — fed by two new `flow_run_*` WS frames on a coder user channel. Qwen-only by design (Claude Code remains the Claude path); the existing model-competition Arena stays a separate feature. The flow launcher and the `/` slash menu both carry chevron-expandable per-item explanations (an always-on one-liner expands to a 1–2 sentence what-it-does / when-to-use blurb, condensed from each Han skill's own description), with a "read-only" pill pinned in the launcher and the fast/concise toggle wired through to the workers. Spec/plan in `openspec/changes/orchestrator`; coder 373 tests green (42 new scheduler/resume/read-only decision tests), contracts/coder/server builds + web tsc clean. Built on `v2.7.16-container-git-safedir`; pairs conceptually with the earlier `v2.7.12-audit-cleanup` multi-agent orchestration.

CLAUDE.md

@@ -74,11 +74,11 @@ Schema CHECK migration order when renaming allowed values: (1) `ALTER TABLE ...
 
 Required: `DATABASE_URL`, `LLAMA_SWAP_URL`. Optional: `PORT` (3000), `HOST` (0.0.0.0), `PROJECT_ROOT_WHITELIST` (/opt, read-only add-existing scope), `BOOTSTRAP_ROOT` (/opt/projects, writable bootstrap mkdir target — host must `mkdir -p` it before container start), `DEFAULT_MODEL`, `LOG_LEVEL`, `SEARXNG_URL` (default `http://100.114.205.53:8888` — internal Tailscale; the public host is behind Authelia, unusable from server context), `BOOCODE_TOOLS` (`core`|`standard`|`all`, default `all`; a ceiling, never expands an agent's whitelist), `MCP_CONFIG_PATH` (default `/data/mcp.json`, opencode `mcpServers` shape; missing = no MCP), `CONTEXT7_API_KEY` (the Context7 MCP key, referenced from `data/mcp.json` as `"{env:CONTEXT7_API_KEY}"`). `data/mcp.json` is **gitignored** but no longer holds secrets — string values support opencode-style `{env:VAR}` substitution (`mcp-config.ts:substituteEnvVars`, applied before Zod validation; unset var → `''` + warn), so real keys live in `.env`; template `data/mcp.example.json`. A config-only edit there needs only `docker compose restart boocode` (data/ is bind-mounted); changing a referenced secret edits `.env`. MCP loads at server startup with per-server graceful degradation; the coder does NOT load MCP (BooChat only).
 
-BooCoder at port 9502: `curl http://100.114.205.53:9502/api/health`. Runs as `boocoder.service` on the host (not Docker). Deploy: `pnpm -C packages/contracts build && pnpm -C apps/server build && pnpm -C apps/coder build && sudo systemctl restart boocoder`. Health reports tool count: `{"ok":true,"db":true,"tools":33}`.
+BooCoder at port 9502: `curl http://100.114.205.53:9502/api/health`. Runs as `boocoder.service` on the host (not Docker). Its env file `apps/coder/.env.host` is gitignored (`.env.*`, with `!.env.example`) — a fresh host recreates it from `.env.example` (incl. `CLAUDE_SDK_BACKEND=1` for the Claude Agent-SDK backend). Deploy: `pnpm -C packages/contracts build && pnpm -C apps/server build && pnpm -C apps/coder build && sudo systemctl restart boocoder`. Health reports tool count: `{"ok":true,"db":true,"tools":33}`.
 
 - `FAST_MODEL` (optional) — cheaper model for titles, summaries, labeling (auto_name.ts, tool-summaries.ts). Falls back to session model or DEFAULT_MODEL. Set to a small llama-swap model (e.g. `nemotron-nano-4b`) to avoid loading the 35B for 20-token calls.
 - Qwen Code dispatch: `OPENAI_BASE_URL=http://100.101.41.16:8401/v1 OPENAI_API_KEY=dummy qwen -p "<task>" --output-format stream-json`. Install: `npm install -g @qwen-code/qwen-code@latest`. Node ≥22 on host (container stays Node 20; BooCoder dispatches via direct spawn on host). No `--yolo` flag — `-p` runs autonomously without prompts. ACP bridge is an HTTP daemon (not stdio); use PTY dispatch.
-- Arena: `POST /api/arena {project_id, input, contestants: [{agent?, model?}]}` dispatches the same task to N models/agents in parallel; each contestant gets its own task + worktree. `GET /api/arena/:id` for results; `POST /api/arena/:id/select/:task_id` picks a winner.
+- Arena: `POST /api/battles {project_id, battle_type, prompt, contestants}` starts a battle; `GET /api/battles/:id` returns battle + contestants + cross-examinations; `POST /api/battles/:id/stop` cancels; `POST /api/battles/:id/analyze` triggers/re-triggers two-stage digest→judge analysis; `GET /api/battles/:id/analysis` reads `analysis.md`; `POST /api/battles/:id/cross-examine {identity, model}` runs a cross-examination. All `/api/battles*` routes are served by `apps/coder` at port 9502 (proxied through `apps/server` as `/api/coder/battles*`).
 
 ## Workflow
 

CONTEXT.md

@@ -0,0 +1,67 @@
+# Context: BooCode
+
+Glossary of the domain language. Terms only — no implementation detail.
+
+## Workspace
+
+- **Pane** — one tile in the multi-pane workspace. Each pane has a *kind*:
+  Chat (BooChat), Coder (BooCoder), Terminal (BooTerm), Orchestrator, Arena,
+  plus artifact/settings kinds.
+
+- **Backend** — an AI engine a task is dispatched to: *native* (BooChat
+  inference on a local llama-swap model) or an *external* CLI agent (Claude Code,
+  OpenCode, Qwen, Goose). Code sometimes calls this the "agent" (`tasks.agent`).
+
+- **BooChat Agent** (a.k.a. *persona*) — a preset from the `data/AGENTS.md`
+  registry (e.g. "Code Reviewer", "Debugger"): a system prompt + tool whitelist +
+  sampling knobs that runs **on the native backend** with a chosen model.
+  Distinct from a Backend — this is the overloaded sense of "agent" the UI's
+  Agent picker selects.
+
+## Arena
+
+A way to run the **same prompt** against several AI competitors at once and pick
+the best result.
+
+- **Battle** — one Arena run. Dated. Produces a results folder at
+  `/<project-root>/Arena/<dated-battle>/`. (The earlier API-only feature called
+  this an "arena"; a Battle is one such run.)
+
+- **Battle Type** — what is being compared:
+  - *Coding* — Contestants change code; a result is the **diff** they produced
+    (plus their explanation). Each Contestant works in its own worktree.
+  - *Q&A* — Contestants answer a prompt; a result is the **text answer**. No
+    code changes.
+
+- **Contestant** — one competitor in a Battle, given the Battle's prompt. What
+  defines a Contestant depends on Battle Type:
+  - *Coding* — a **Backend + Model** (e.g. Claude Code + opus, native BooCode +
+    35b). Each works in its own isolated git **worktree** (a branched on-disk
+    copy of the project). Contestants do not see each other's work.
+  - *Q&A* — a **BooChat Agent (persona) + Model** (e.g. Debugger + 35b), running
+    on the native backend only. No worktree (no code changes).
+  The same model can appear under two Contestants, so a Contestant's identity is
+  the (backend-or-persona, model) pair, not the model alone.
+
+- **Benchmark** — per-Contestant performance captured during a Battle. Wall-clock
+  **duration** is recorded for every Contestant; **throughput** (tokens/sec) is
+  recorded only for local (llama-swap) models, which are the ones the speed
+  comparison is meaningful for.
+
+- **Arena results folder** (`/<project-root>/Arena/<dated-battle>/`) — where a
+  Battle's *results* are written (not the working copies — those stay in each
+  Contestant's worktree). Holds the per-Contestant result and the final
+  analysis.
+
+- **Lane** — how a Battle's Contestants are scheduled. The *local lane* holds
+  every llama-swap-backed Contestant and runs them strictly one at a time (the
+  local server can only load one model at a time, which also keeps their speed
+  Benchmark fair). The *cloud lane* holds cloud-backed Contestants (Claude Code,
+  OpenCode-on-cloud) and runs them all in parallel. The two lanes run
+  concurrently with each other.
+
+- **Analysis** — an end-of-Battle judgement of the Contestants' results,
+  produced by the default BooChat model, naming a **Winner**.
+
+- **Cross-examination** — an after-the-Battle step where a chosen model (from any
+  agent) is pointed at the Battle's results to interrogate / compare them.

CURRENT.md

@@ -1,9 +1,9 @@
 # Current focus
 
-Last updated: 2026-06-02
+Last updated: 2026-06-07
 
-- **Last shipped:** `v2.7.8-ember-coder-tabs-model-chips` (2026-06-01)
+- **Last shipped:** `v2.8.0-fork-lifts` (2026-06-07) — eight fork-lift integrations from `/opt/forks`: boocontext sidecar, LSP code intelligence, DCP clean-room pruning, institutional memory, subagent protocol, plugin hook host, inference reliability (tool-shim + loop detectors), and TokenScope token breakdown. Backfills edit safety guards and TokenScope analyzer/persist module.
-- **Branch:** `codebase-audit-cleanup` (audit + cleanup epic, off main HEAD)
+- **Branch:** `main`
-- **In progress:** Phase 3 — stale comments + docs refresh
+- **In progress:** nothing committed — all phases 3-9 of fork-lifts-mit epic are shipped. Optional/exploratory: verify-gate ensembler over pending changes; web Arena token UI display.
 
 See `CHANGELOG.md` for the full shipped history. That file is always authoritative; this file is a quick orientation pointer only.

README.md

@@ -1,10 +1,10 @@
 # boocode
 
-Self-hosted single-user developer chat app. 3-app monorepo: BooChat (read-only chat), BooCoder (write tools + agent dispatch), BooTerm (PTY terminals).
+Self-hosted single-user developer chat app. 3-app monorepo: BooChat (read-only chat), BooCoder (write tools + agent dispatch), BooTerm (PTY terminals) — plus the in-app **Orchestrator**, a deterministic multi-agent conductor that runs read-only Han analysis/review flows on local Qwen.
 
-**Latest release:** `v2.2.1-pane-scoped-chats` (2026-05-26) · [`CHANGELOG.md`](CHANGELOG.md) · **Current focus:** [`CURRENT.md`](CURRENT.md)
+**Latest release:** `v2.7.17-orchestrator` (2026-06-03) · [`CHANGELOG.md`](CHANGELOG.md) · **Current focus:** [`CURRENT.md`](CURRENT.md)
 
-**Agent navigation:** [`AGENTS.md`](AGENTS.md) · **Architecture:** [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) · **Engineering reference:** [`CLAUDE.md`](CLAUDE.md)
+**Architecture:** [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) · **Engineering reference:** [`CLAUDE.md`](CLAUDE.md) · **Roadmap:** [`boocode_roadmap.md`](boocode_roadmap.md)
 
 ## Stack
 
@@ -75,15 +75,16 @@ curl http://100.114.205.53:9502/api/health
 
 ## What's shipped
 
-See [`boocode_roadmap.md`](boocode_roadmap.md) for full version history. Highlights as of **v2.2.1**:
+See [`boocode_roadmap.md`](boocode_roadmap.md) and [`CHANGELOG.md`](CHANGELOG.md) for full version history. Highlights as of **v2.7.17**:
 
-- **BooChat**: streaming chat, file-read tools, compaction, reasoning support, HTML/Markdown artifact panes, cross-repo read grants, MCP client (multi-server + stdio), tool-cost tracking, skills system, builtin agent registry, multi-pane workspace (chat / terminal / coder)
+- **BooChat**: streaming chat, file-read tools, compaction, reasoning support, HTML/Markdown artifact panes, cross-repo read grants, MCP client (multi-server + stdio), tool-cost tracking, skills system, builtin agent registry, multi-pane workspace (chat / terminal / coder / orchestrator)
 - **BooTerm**: in-browser terminal panes via tmux + xterm.js, per-session tmux sessions, SSH-out support
-- **BooCoder (v2.2)**: write tools (`edit_file`, `create_file`, `delete_file`, `apply_pending`, `rewind`), pending-changes queue with diff UI, Paseo-style provider snapshot (7 providers: boocode, cursor, claude, opencode, goose, qwen, copilot), `AgentComposerBar` (provider / mode / model / thinking), ACP dispatch with inline permission prompts + tool/reasoning streaming, PTY fallback, Arena, MCP server (6 tools, stdio), CLI client, human inbox, Boomerang orchestration, path-guard fuzz suite, **pane-scoped chats** (v2.2.1 — each coder/terminal pane owns its chat)
+- **BooCoder**: write tools (`edit_file` with fuzzy matching, `create_file`, `delete_file`, `apply_pending`, `rewind`, git-ref checkpoints), pending-changes queue + a **Files/Git diff panel** (stage / commit / discard), provider snapshot (5 providers: boocode, claude, opencode, goose, qwen — cursor/copilot retired), `AgentComposerBar`, warm ACP + **persistent agent sessions** (opencode HTTP server; claude via the Agent SDK with native session resume) + PTY fallback, config-backed provider lifecycle, Arena (same task → N models), MCP server, CLI client, human inbox, Boomerang orchestration, pane-scoped chats
+- **Orchestrator** (v2.7.17): launch any of 22 read-only Han flows (research, code-review, investigate, architectural-analysis, …) from BooChat or BooCoder via the Workflow button, a slash command, or **+ menu → New Orchestrator**; each step runs as a bounded agent on local Qwen (hard read-only via `qwen --approval-mode plan`), streaming live in a Paseo-style run pane with an evidence-disciplined, adversarially-validated report. Persisted + resumable. `@boocode/contracts` single-sources the cross-app wire contracts (v2.7.13).
 
 ## Planned
 
-- **v2.3 provider lifecycle** — config-backed provider registry (`/data/coder-providers.json`), enable/disable toggles, two-tier probe (openspec drafted). See [`CURRENT.md`](CURRENT.md).
+Most prior roadmap milestones have shipped (see [`boocode_roadmap.md`](boocode_roadmap.md)). What remains is optional/exploratory — e.g. a verify-gate ensembler over pending changes (majority-vote diff ranking). No committed milestones currently in flight.
 
 ## License
 

apps/coder/.env.host

@@ -1,17 +0,0 @@
-NODE_ENV=production
-PORT=9502
-HOST=100.114.205.53
-DATABASE_URL=postgres://boocode:devpass@127.0.0.1:5500/boochat
-LLAMA_SWAP_URL=http://100.101.41.16:8401
-PROJECT_ROOT_WHITELIST=/opt
-BOOTSTRAP_ROOT=/opt/projects
-DEFAULT_MODEL=qwen3.6-35b-a3b-mxfp4
-LOG_LEVEL=info
-SEARXNG_URL=http://100.114.205.53:8888
-GITEA_BASE_URL=https://git.indifferentketchup.com
-GITEA_USER=indifferentketchup
-GITEA_SSH_HOST=100.114.205.53:2222
-MCP_CONFIG_PATH=/data/mcp.json
-SKILLS_ROOT=/opt/boocode/data/skills
-CODER_PROVIDERS_PATH=/opt/boocode/data/coder-providers.json
-CLAUDE_SDK_BACKEND=1

apps/coder/CLAUDE.md

@@ -32,3 +32,15 @@
 - **Claude SDK backend tool RESULTS arrive as `type:'user'` SDK messages** (tool_result content blocks): `mapSdkMessage` (`claude-sdk-map.ts`) MUST map the `user` case → a terminal `tool_update` (completed/failed + output), else the tool_call persists `status:'running'` and the UI spinner never stops. The dispatcher's `tool_update` path then publishes + persists it.
 - **ACP command discovery is async**: `acp-probe.ts` must poll after `newSession` for `available_commands_update` (commands arrive in a later notification; reading synchronously captures 0). PTY providers (claude) discover from disk via `claude-command-discovery.ts` (`~/.claude/commands` + `enabledPlugins`, bare names, deduped). `AgentCommand.kind` tags `'command'` vs `'skill'`; `CoderPane`'s `slashGroups` splits them into icon'd groups. `SlashCommandPicker`'s `groups?` prop is opt-in.
 - **A new per-message coder field silently drops unless you update every mapper**: the HTTP read SELECT + `mapCoderMessageRow` (`apps/coder/src/routes/messages.ts`), **the WS `snapshot` SELECT (`apps/coder/src/routes/ws.ts`)** — it has its OWN column list and the client's `snapshot` handler `setMessages`-overwrites the HTTP load, so a field present in the HTTP route but absent here shows live yet vanishes on refresh — `CoderPane.tsx` (`RawCoderMessage`/`CoderMessage`/`mapCoderTimelineRow` + the live `message_complete` WS reducer), `CoderMessageWire` (`CoderMessageList.tsx`), and `api/types.ts`. The client `mapCoderTimelineRow` whitelists fields — easiest to forget. This bit `model` twice: the client chain (`v2.7.9`) and then the WS snapshot SELECT (`v2.7.11`) — the chip showed live but vanished on coder refresh until both were fixed.
+
+## Orchestrator (v2.7.17)
+
+- **In-app multi-agent conductor**: `services/flow-runner.ts` runs a flow by inserting each step as a `tasks` row (the existing dispatcher runs it) and advancing on a new `onTaskTerminal` dispatcher-deps hook; persisted in `flow_runs`/`flow_steps` (resumed at startup via `initResume`). The 22 conductor flow defs + Spine factory are re-homed under `src/conductor/`. Pure scheduler/resume helpers in `flow-runner-decisions.ts`. Full design: `openspec/changes/archived/orchestrator/`.
+- **Read-only is load-bearing — don't add a dispatch path that bypasses it.** Every step dispatches `agent='qwen', mode_id='plan'`; `dispatcher.ts` force-routes qwen+plan to the PTY `--approval-mode plan` gate and HARD-FAILS the task (never falls to write-capable native inference) when qwen is unavailable (`shouldFailOnMissingAgent`). `BOOCODE_TOOLS` gates BooChat's NATIVE inference tools only — it does NOT govern an external CLI agent (qwen/opencode bring their own write tools); read-only for a dispatched agent is the agent-layer mode (PTY `--approval-mode plan`; ACP `setSessionMode` is fail-OPEN by default, fail-CLOSED for `plan` via `READ_ONLY_MODE_IDS` in `acp-dispatch.ts`).
+
+## Edit safety guards (v2.8)
+
+- **`services/edit-guards.ts`** — `validateEditResult(original, updated, filePath)` runs in `pending_changes.ts` immediately before `writeFileAtomic`. Rejects catastrophic truncation (>60% char loss AND >50% line loss). Throws a `formatGuardError` message that percolates to the agent as a visible error.
+- **`services/edit-guards-imports.ts`** — `checkDroppedImports(original, updated, filePath)` detects removed import/require lines. Called alongside the truncation guard.
+- Both guards run on the `/apply` path only (not on queue). Re-queued identical edits re-validate at apply time.
+- Guard functions are pure — no DB or filesystem access. Easy to unit-test.

apps/coder/src/conductor/flows/index.ts

@@ -24,6 +24,7 @@ import {
 } from './planning.js';
 import { adr, codingStandard, runbook, tdd, stakeholderSummary } from './authoring.js';
 import { codeReview } from './code-review.js';
+import { parallelResearch } from './parallel-research.js';
 
 const spines: Spine[] = [
   // analysis / research
@@ -53,7 +54,7 @@ const spines: Spine[] = [
   stakeholderSummary,
 ];
 
-const bespoke: Flow[] = [codeReview];
+const bespoke: Flow[] = [codeReview, parallelResearch];
 
 const ALL: Flow[] = [...spines.map(buildSpineFlow), ...bespoke];
 

apps/coder/src/conductor/flows/parallel-research.ts

@@ -0,0 +1,59 @@
+import type { Flow, Step, StepContext } from '../types.js';
+
+const q = (ctx: StepContext) => String(ctx.input.question);
+
+/**
+ * Parallel research flow — dispatches 3 research agents simultaneously,
+ * then synthesizes the result on the first one to complete.
+ */
+export const parallelResearch: Flow = {
+  name: 'parallel-research',
+  description: 'Research from 3 angles in parallel, synthesize results on first completion',
+  steps: [
+    {
+      id: 'angle-web',
+      kind: 'agent',
+      agent: 'research-analyst',
+      run: (ctx) =>
+        `Research the following question from a web / prior-art perspective:\n\n${q(ctx)}`,
+    },
+    {
+      id: 'angle-code',
+      kind: 'agent',
+      agent: 'codebase-explorer',
+      deps: [],
+      run: (ctx) =>
+        `Research the following question from a codebase analysis perspective:\n\n${q(ctx)}`,
+    },
+    {
+      id: 'angle-security',
+      kind: 'agent',
+      agent: 'adversarial-security-analyst',
+      deps: [],
+      run: (ctx) =>
+        `Research the following question from a security perspective:\n\n${q(ctx)}`,
+    },
+    {
+      id: 'synthesize',
+      kind: 'code',
+      deps: ['angle-web', 'angle-code', 'angle-security'],
+      trigger_rule: 'one_success',
+      run: (ctx) => {
+        const web = ctx.results['angle-web'];
+        const code = ctx.results['angle-code'];
+        const security = ctx.results['angle-security'];
+        const parts = [
+          '# Parallel Research Synthesis',
+          '',
+          web ? `## Web Angle\n${web}` : '## Web Angle\n*(not yet completed)*',
+          code ? `## Code Angle\n${code}` : '## Code Angle\n*(not yet completed)*',
+          security ? `## Security Angle\n${security}` : '## Security Angle\n*(not yet completed)*',
+        ];
+        return parts.join('\n\n');
+      },
+    },
+  ],
+  render: (ctx) => {
+    return ctx.results['synthesize'] ?? 'No synthesis produced.';
+  },
+};

apps/coder/src/conductor/types.ts

@@ -38,7 +38,9 @@ export interface StepContext {
   readonly model?: string;
 }
 
-export type StepKind = 'agent' | 'code';
+export type StepKind = 'agent' | 'code' | 'approval';
+
+export type TriggerRule = 'all_success' | 'one_success' | 'all_done';
 
 export interface Step {
   /** unique id within the flow; other steps depend on it by this id */
@@ -46,6 +48,8 @@ export interface Step {
   kind: StepKind;
   /** ids that must complete (or skip) before this step runs */
   deps?: string[];
+  /** how dependency satisfaction is evaluated (default: all_success) */
+  trigger_rule?: TriggerRule;
   /** for kind:'agent' — the persona file name under conductor/agents (no .md) */
   agent?: string;
   /**

apps/coder/src/index.ts

@@ -13,7 +13,7 @@ import type { WsFrame } from '@boocode/contracts/ws-frames';
 // v2.0.0 Phase 2C: write tools + adapter for BooChat ToolDef compatibility.
 import { WRITE_TOOLS } from './services/tools/index.js';
 import { adaptWriteTool } from './services/tools/adapter.js';
-import { setInferenceContext, clearInferenceContext } from './services/tools/inference_context.js';
+import { runWithInferenceContext } from './services/tools/inference_context.js';
 // Routes
 import { registerMessageRoutes } from './routes/messages.js';
 import { registerSkillRoutes } from './routes/skills.js';
@@ -23,8 +23,8 @@ import { registerAgentSessionRoutes } from './routes/agent-sessions.js';
 import { registerTaskRoutes } from './routes/tasks.js';
 import { registerInboxRoutes } from './routes/inbox.js';
 import { registerStatsRoutes } from './routes/stats.js';
-import { registerArenaRoutes } from './routes/arena.js';
 import { registerRunsRoutes } from './routes/runs.js';
+import { registerArenaRoutes } from './routes/arena.js';
 import { registerProviderRoutes } from './routes/providers.js';
 import { registerWorktreeSafetyRoutes } from './routes/worktree-safety.js';
 import { registerLifecycleRoutes } from './routes/lifecycle.js';
@@ -34,10 +34,13 @@ import { createDispatcher } from './services/dispatcher.js';
 // Orchestrator (Phase 2): DB-backed flow-runner; advances on the dispatcher's
 // onTaskTerminal hook.
 import { createFlowRunner } from './services/flow-runner.js';
+// Arena: DB-backed battle-runner; also advances on the onTaskTerminal hook.
+import { createBattleRunner, type DispatchContestantFn } from './services/arena-runner.js';
+import { createAnalyzer } from './services/arena-analyzer.js';
 import { agentPool } from './services/agent-pool.js';
 import { createOrphanWorktreeReaper } from './services/orphan-worktree-reaper.js';
 import { probeAgents } from './services/agent-probe.js';
-import { getProviderSnapshot, persistProbedModels } from './services/provider-snapshot.js';
+import { getProviderSnapshot, persistProbedModels, fetchLlamaSwapModels } from './services/provider-snapshot.js';
 import { setPermissionHooks } from './services/permission-waiter.js';
 import { publishAgentStatus } from './services/agent-status-publish.js';
 import { homedir } from 'node:os';
@@ -171,22 +174,27 @@ async function main() {
     }
   );
 
-  // Wrap the inference runner to set/clear the write-tool context around each run.
+  // Wrap the inference runner to bind the write-tool context around each run.
-  // The inference runner calls enqueue() which fires asynchronously — we hook
+  // enqueue() starts its async loop synchronously, so wrapping the call in
-  // into the enqueue to set context before the run starts.
+  // runWithInferenceContext propagates the per-run context (sql, sessionId, the
+  // Plan/Ask/Bypass gate) through every awaited tool execution — and concurrent
+  // runs (a user message racing a dispatcher-polled native task) each get their
+  // own, instead of clobbering a shared global.
   const inferenceApi = {
-    enqueue: (sessionId: string, chatId: string, assistantId: string, user: string) => {
+    enqueue: (
-      // Set the inference context so write tools can access sql + sessionId.
+      sessionId: string,
-      // The context persists for the duration of the inference run. Since
+      chatId: string,
-      // BooCoder is single-user and runs one inference at a time per session,
+      assistantId: string,
-      // this module-level state is safe.
+      user: string,
-      setInferenceContext({ sql, sessionId, taskId: null });
+      permissionMode?: 'plan' | 'ask' | 'bypass',
-      inference.enqueue(sessionId, chatId, assistantId, user);
+    ) => {
+      runWithInferenceContext({ sql, sessionId, taskId: null, permissionMode }, () => {
+        inference.enqueue(sessionId, chatId, assistantId, user);
+      });
     },
     cancel: async (sessionId: string, chatId: string) => {
-      const result = await inference.cancel(sessionId, chatId);
+      // No context to clear — AsyncLocalStorage scopes it to each run's own chain.
-      clearInferenceContext();
+      return inference.cancel(sessionId, chatId);
-      return result;
     },
     hasActive: (chatId: string) => inference.hasActive(chatId),
   };
@@ -220,31 +228,119 @@ async function main() {
 
   // Orchestrator (Phase 2): the flow-runner reacts to the dispatcher's
   // onTaskTerminal hook to advance flow_runs. Created before the dispatcher so its
-  // terminal callback can be wired in. Its launch() is driven by the runs route
+  // terminal callback can be wired in.
-  // (a later phase); resume on startup is a later phase too.
   const flowRunner = createFlowRunner({ sql, broker, log: app.log, config });
 
-  // Phase 4: dispatcher — polls tasks table and runs inference. onTaskTerminal
+  // Arena SEAM (a): build the local-model set from the live llama-swap model list.
-  // notifies the flow-runner when a step's task settles (D-2).
+  // Both bare IDs ('qwen3.6-35b') and prefixed IDs ('llama-swap/qwen3.6-35b') are
+  // included so opencode-style prefixed contestants and native-style bare contestants
+  // both classify correctly as local.
+  const localModelsList = await fetchLlamaSwapModels(config).catch(() => []);
+  const localModels = new Set([
+    ...localModelsList.map((m) => m.id),
+    ...localModelsList.map((m) => `llama-swap/${m.id}`),
+  ]);
+
+  // Arena dispatch function — Phase 4 SEAM (b).
+  // Coding: insert a tasks row with agent=identity (null for native/boocode);
+  //   the dispatcher creates a worktree and runs the external agent (or native).
+  // Q&A: pre-create a session with agent_id stamped to the persona slug so native
+  //   inference loads the persona's system_prompt + tools from AGENTS.md;
+  //   task.session_id is pre-set so runNativeInference reuses the session.
+  const dispatchContestant: DispatchContestantFn = async ({
+    projectId,
+    prompt,
+    identity,
+    model,
+    battleType,
+  }) => {
+    if (battleType === 'qa') {
+      const sessionName = `Arena Q&A [${identity}]: ${prompt.slice(0, 30)}`;
+      const [session] = await sql<{ id: string }[]>`
+        INSERT INTO sessions (project_id, name, model, agent_id, status)
+        VALUES (${projectId}, ${sessionName}, ${model}, ${identity}, 'open')
+        RETURNING id
+      `;
+      const [task] = await sql<{ id: string }[]>`
+        INSERT INTO tasks (project_id, input, model, session_id)
+        VALUES (${projectId}, ${prompt}, ${model}, ${session!.id})
+        RETURNING id
+      `;
+      return { taskId: task!.id, sessionId: session!.id };
+    }
+    // Coding: boocode = native inference (no external agent); any other identity
+    // is an external agent name (claude, opencode, qwen, goose) that maps to
+    // available_agents and gets its own per-task worktree via runExternalAgent.
+    // Session is created lazily by the dispatcher, so sessionId is unknown here.
+    const agentName = identity === 'boocode' ? null : identity;
+    const [task] = await sql<{ id: string }[]>`
+      INSERT INTO tasks (project_id, input, agent, model)
+      VALUES (${projectId}, ${prompt}, ${agentName}, ${model})
+      RETURNING id
+    `;
+    return { taskId: task!.id, sessionId: null };
+  };
+
+  // Arena analyzer: two-stage digest→judge (v1). Pluggable seam — a v2 Han
+  // Orchestrator flow can replace this without schema changes.
+  const analyzer = createAnalyzer({
+    sql,
+    broker,
+    log: app.log,
+    config,
+    localModels,
+  });
+
+  // Arena battle-runner: notified on the same onTaskTerminal hook as the flow-runner.
+  const battleRunner = createBattleRunner({
+    sql,
+    broker,
+    log: app.log,
+    dispatch: dispatchContestant,
+    onBattleComplete: (battleId) => {
+      void analyzer.analyze(battleId);
+    },
+    onCrossExamStart: ({ battleId, crossExamId, identity, model }) => {
+      void analyzer.crossExamine(battleId, crossExamId, { identity, model });
+    },
+    localModels,
+  });
+
+  // Compose onTaskTerminal: both flow-runner and battle-runner are notified.
+  // Each ignores tasks it doesn't own (flow-runner checks flow_steps.task_id;
+  // battle-runner checks contestants.task_id).
+  const onTaskTerminal = (taskId: string, state: string): void => {
+    flowRunner.handleTaskTerminal(taskId, state);
+    battleRunner.handleTaskTerminal(taskId, state);
+  };
+
+  // Phase 4: dispatcher — polls tasks table and runs inference. The composed
+  // onTaskTerminal hook notifies both the flow-runner and the battle-runner when
+  // any task settles.
   const dispatcher = createDispatcher({
     sql,
     inference: inferenceApi,
     broker,
     log: app.log,
     config,
-    onTaskTerminal: flowRunner.handleTaskTerminal,
+    onTaskTerminal,
   });
   dispatcher.start();
 
-  // Phase 5: re-advance any flow_runs that were 'running' when the service last
+  // Re-advance in-flight flow_runs and battles after a coder restart. Both run
-  // stopped (D-9). Runs AFTER dispatcher.start() so re-dispatched 'pending' tasks
+  // AFTER dispatcher.start() so re-dispatched 'pending' tasks are picked up.
-  // are picked up by the dispatcher's startup poll.
   void flowRunner.initResume().catch((err) => {
     app.log.error(
       { err: err instanceof Error ? err.message : String(err) },
       'flow-runner: initResume failed',
     );
   });
+  void battleRunner.initResume().catch((err) => {
+    app.log.error(
+      { err: err instanceof Error ? err.message : String(err) },
+      'arena: initResume failed',
+    );
+  });
 
   // v2.6 Phase 3: configure + start the agent-pool lifecycle sweep (idle-TTL +
   // LRU-cap eviction of warm backends, plus each backend's proactive health probe)
@@ -281,8 +377,8 @@ async function main() {
   registerTaskRoutes(app, sql, inferenceApi, dispatcher.cancelExternalTask);
   registerInboxRoutes(app, sql);
   registerStatsRoutes(app, sql);
-  registerArenaRoutes(app, sql);
   registerRunsRoutes(app, sql, flowRunner, dispatcher.cancelExternalTask);
+  registerArenaRoutes(app, sql, battleRunner, dispatcher.cancelExternalTask, config);
   registerProviderRoutes(app, sql, config);
   registerWorktreeSafetyRoutes(app, sql);
   registerLifecycleRoutes(app, sql);

apps/coder/src/plugins/host.ts

@@ -0,0 +1,42 @@
+export type HookName =
+  | 'tool.execute.before'
+  | 'tool.execute.after'
+  | 'turn.start'
+  | 'turn.end'
+  | 'task.terminal';
+
+export interface ToolHookContext {
+  tool: string;
+  args: Record<string, unknown>;
+  projectRoot: string;
+  sessionId: string;
+}
+
+export interface ToolResultContext extends ToolHookContext {
+  result: unknown;
+}
+
+export type PluginHook = (ctx: any) => Promise<any>;
+
+const hooks = new Map<HookName, PluginHook[]>();
+
+export function registerHook(name: HookName, fn: PluginHook): void {
+  const list = hooks.get(name) || [];
+  list.push(fn);
+  hooks.set(name, list);
+}
+
+export async function emitHook(name: HookName, ctx: any): Promise<any> {
+  const list = hooks.get(name);
+  if (!list) return ctx;
+  let current = ctx;
+  for (const fn of list) {
+    const result = await fn(current);
+    if (result !== undefined) current = result;
+  }
+  return current;
+}
+
+export function clearHooks(): void {
+  hooks.clear();
+}

apps/coder/src/routes/arena.ts

@@ -1,136 +1,412 @@
 /**
- * v2.0.5: Arena routes — competitive dispatch of the same task to multiple agents.
+ * Arena routes — HTTP surface for the Battle UI.
  *
- * POST /api/arena        — create an arena with 2-5 contestants
+ * POST /api/battles                         — launch a battle
- * GET  /api/arena/:id    — get all tasks in an arena
+ * GET  /api/battles?project_id=             — list battles for a project
- * POST /api/arena/:id/select/:task_id — mark a task as the arena winner
+ * GET  /api/battles/:id                     — one battle + contestants + cross-exams
+ * POST /api/battles/:id/stop                — cancel a running battle
+ * POST /api/battles/:id/analyze             — trigger analysis (Phase 5 fills the logic)
+ * POST /api/battles/:id/cross-examine       — start a cross-examination (Phase 5 fills the logic)
+ *
+ * Mirrors the shape of runs.ts (Orchestrator routes). Battle creation delegates to
+ * the battle-runner; cancellation calls cancelBattle then aborts in-flight tasks
+ * via the dispatcher's cancelExternalTask.
  */
 import type { FastifyInstance } from 'fastify';
 import { z } from 'zod';
+import { readFile } from 'node:fs/promises';
+import { join } from 'node:path';
 import type { Sql } from '../db.js';
+import type { Config } from '../config.js';
+import type { BattleRunner } from '../services/arena-runner.js';
+import type { ExternalCancelFn } from './tasks.js';
+import { arenaModelCall } from '../services/arena-model-call.js';
 
-const ContestantSchema = z.object({
+// ─── Validation schemas ───────────────────────────────────────────────────────
-  agent: z.string().max(100).optional(),
+
-  model: z.string().max(200).optional(),
+const UuidParam = z.string().uuid();
-  mode_id: z.string().max(200).optional(),
+
-  thinking_option_id: z.string().max(200).optional(),
+const ContestantInput = z.object({
+  identity: z.string().min(1).max(200),
+  model: z.string().min(1).max(200),
 });
 
-const CreateArenaBody = z.object({
+const CreateBattleBody = z.object({
   project_id: z.string().uuid(),
-  input: z.string().min(1).max(64_000),
+  battle_type: z.enum(['coding', 'qa']),
-  contestants: z.array(ContestantSchema).min(2).max(5),
+  prompt: z.string().min(1).max(64_000),
+  contestants: z
+    .array(ContestantInput)
+    .min(2, 'at least 2 contestants required')
+    .max(6, 'at most 6 contestants allowed'),
 });
 
-interface TaskRow {
+const ListBattlesQuery = z.object({
-  id: string;
+  project_id: z.string().uuid(),
-  agent: string | null;
+});
-  model: string | null;
-  mode_id: string | null;
-  thinking_option_id: string | null;
-  state: string;
-}
 
-export function registerArenaRoutes(app: FastifyInstance, sql: Sql): void {
+const CrossExamineBody = z.object({
-  // POST /api/arena — create a new arena
+  identity: z.string().min(1).max(200),
-  app.post('/api/arena', async (req, reply) => {
+  model: z.string().min(1).max(200),
-    const parsed = CreateArenaBody.safeParse(req.body);
+});
+
+const SetWinnerBody = z.object({
+  winner_contestant_id: z.string().uuid().nullable(),
+});
+
+// ─── Route registration ───────────────────────────────────────────────────────
+
+const GeneratePromptBody = z.object({
+  description: z.string().min(1).max(2_000),
+});
+
+export function registerArenaRoutes(
+  app: FastifyInstance,
+  sql: Sql,
+  battleRunner: BattleRunner,
+  cancelExternal: ExternalCancelFn,
+  config: Config,
+): void {
+
+  // POST /api/battles/generate-prompt — draft a fuller battle prompt from a
+  // short description using the default BooChat model. One-shot, non-streaming.
+  // Must be registered BEFORE /api/battles/:id so the literal 'generate-prompt'
+  // path is not mistaken for a UUID param.
+  app.post('/api/battles/generate-prompt', async (req, reply) => {
+    const parsed = GeneratePromptBody.safeParse(req.body);
     if (!parsed.success) {
       reply.code(400);
       return { error: 'invalid body', details: parsed.error.flatten() };
     }
 
-    const { project_id, input, contestants } = parsed.data;
+    const { description } = parsed.data;
-    const arenaId = crypto.randomUUID();
 
-    const tasks: TaskRow[] = [];
+    try {
-    for (const contestant of contestants) {
+      const prompt = await arenaModelCall({
-      const [task] = await sql<TaskRow[]>`
+        config,
-        INSERT INTO tasks (project_id, input, agent, model, mode_id, thinking_option_id, arena_id)
+        model: config.DEFAULT_MODEL,
-        VALUES (
+        system: [
-          ${project_id},
+          'You are a battle-prompt writer for an AI Arena.',
-          ${input},
+          'The user gives you a short description of a coding or Q&A challenge.',
-          ${contestant.agent ?? null},
+          'Expand it into a clear, self-contained prompt (2–6 sentences) that any AI model can act on.',
-          ${contestant.model ?? null},
+          'Include specific acceptance criteria where helpful.',
-          ${contestant.mode_id ?? null},
+          'Output ONLY the prompt — no preamble, no labels, no meta-commentary.',
-          ${contestant.thinking_option_id ?? null},
+        ].join(' '),
-          ${arenaId}
+        user: description,
-        )
+        maxTokens: 400,
-        RETURNING id, agent, model, mode_id, thinking_option_id, state
+        temperature: 0.6,
-      `;
+      });
-      tasks.push(task!);
+      return { prompt };
+    } catch (err) {
+      app.log.warn(
+        { err: err instanceof Error ? err.message : String(err) },
+        'arena generate-prompt: model call failed',
+      );
+      reply.code(502);
+      return { error: 'model call failed' };
     }
+  });
+
+  // POST /api/battles — launch a battle
+  app.post('/api/battles', async (req, reply) => {
+    const parsed = CreateBattleBody.safeParse(req.body);
+    if (!parsed.success) {
+      reply.code(400);
+      return { error: 'invalid body', details: parsed.error.flatten() };
+    }
+
+    const { project_id, battle_type, prompt, contestants } = parsed.data;
+
+    // Reject duplicate (identity, model) pairs up front — the schema UNIQUE
+    // constraint would catch it too, but an early 422 is friendlier.
+    const seen = new Set<string>();
+    for (const c of contestants) {
+      const key = `${c.identity}::${c.model}`;
+      if (seen.has(key)) {
+        reply.code(422);
+        return {
+          error: 'duplicate_contestant',
+          message: `duplicate contestant: identity="${c.identity}" model="${c.model}"`,
+        };
+      }
+      seen.add(key);
+    }
+
+    // Verify project exists
+    const [proj] = await sql<{ id: string }[]>`SELECT id FROM projects WHERE id = ${project_id}`;
+    if (!proj) {
+      reply.code(404);
+      return { error: 'project not found' };
+    }
+
+    const { battleId } = await battleRunner.startBattle({
+      projectId: project_id,
+      battleType: battle_type,
+      prompt,
+      contestants,
+    });
 
     reply.code(201);
-    return {
+    return { battle_id: battleId };
-      arena_id: arenaId,
-      tasks: tasks.map((t) => ({
-        id: t.id,
-        agent: t.agent,
-        model: t.model,
-        mode_id: t.mode_id,
-        thinking_option_id: t.thinking_option_id,
-        state: t.state,
-      })),
-    };
   });
 
-  // GET /api/arena/:arena_id — list all tasks in an arena
+  // GET /api/battles?project_id= — list battles, most-recent-first
-  app.get<{ Params: { arena_id: string } }>('/api/arena/:arena_id', async (req, reply) => {
+  app.get('/api/battles', async (req, reply) => {
-    const { arena_id } = req.params;
+    const parsed = ListBattlesQuery.safeParse(req.query);
-
+    if (!parsed.success) {
-    // Validate UUID format
-    const uuidRegex = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
-    if (!uuidRegex.test(arena_id)) {
       reply.code(400);
-      return { error: 'invalid arena_id format' };
+      return { error: 'invalid query', details: parsed.error.flatten() };
     }
 
-    const tasks = await sql`
+    const battles = await sql`
-      SELECT id, project_id, state, input, output_summary, agent, model, mode_id, thinking_option_id, execution_path, session_id, started_at, ended_at, created_at, arena_id
+      SELECT id, project_id, battle_type, prompt, status,
-      FROM tasks
+             winner_contestant_id, results_path, error,
-      WHERE arena_id = ${arena_id}
+             created_at, updated_at
-      ORDER BY created_at
+      FROM battles
+      WHERE project_id = ${parsed.data.project_id}
+      ORDER BY created_at DESC
+      LIMIT 100
     `;
 
-    if (tasks.length === 0) {
+    return { battles };
-      reply.code(404);
-      return { error: 'arena not found' };
-    }
-
-    return { arena_id, tasks };
   });
 
-  // POST /api/arena/:arena_id/select/:task_id — mark the winner
+  // GET /api/battles/:id — one battle + its contestants + cross-examinations
-  app.post<{ Params: { arena_id: string; task_id: string } }>(
+  app.get<{ Params: { id: string } }>('/api/battles/:id', async (req, reply) => {
-    '/api/arena/:arena_id/select/:task_id',
+    const parsedId = UuidParam.safeParse(req.params.id);
-    async (req, reply) => {
+    if (!parsedId.success) {
-      const { arena_id, task_id } = req.params;
+      reply.code(400);
-
+      return { error: 'invalid id' };
-      // Verify the task belongs to this arena
-      const rows = await sql<{ id: string; state: string; arena_id: string | null }[]>`
-        SELECT id, state, arena_id FROM tasks WHERE id = ${task_id}
-      `;
-
-      if (rows.length === 0) {
-        reply.code(404);
-        return { error: 'task not found' };
-      }
-
-      const task = rows[0]!;
-      if (task.arena_id !== arena_id) {
-        reply.code(409);
-        return { error: 'task does not belong to this arena' };
-      }
-
-      // Mark as selected via output_summary prefix (lightweight — no schema change)
-      await sql`
-        UPDATE tasks
-        SET output_summary = COALESCE('[SELECTED] ' || output_summary, '[SELECTED]')
-        WHERE id = ${task_id}
-      `;
-
-      return { selected: true, task_id, arena_id };
     }
-  );
+    const id = parsedId.data;
+
+    const [battle] = await sql<{
+      id: string;
+      project_id: string;
+      battle_type: string;
+      prompt: string;
+      status: string;
+      winner_contestant_id: string | null;
+      results_path: string | null;
+      error: string | null;
+      created_at: unknown;
+      updated_at: unknown;
+    }[]>`
+      SELECT id, project_id, battle_type, prompt, status,
+             winner_contestant_id, results_path, error,
+             created_at, updated_at
+      FROM battles WHERE id = ${id}
+    `;
+
+    if (!battle) {
+      reply.code(404);
+      return { error: 'battle not found' };
+    }
+
+    const contestants = await sql`
+      SELECT id, battle_id, identity, model, lane, task_id, worktree_id,
+             status, duration_ms, tokens_per_sec, cost_tokens, token_breakdown, result_path, error,
+             created_at, updated_at
+      FROM contestants
+      WHERE battle_id = ${id}
+      ORDER BY created_at ASC
+    `;
+
+    const crossExaminations = await sql`
+      SELECT id, battle_id, identity, model, verdict, created_at
+      FROM cross_examinations
+      WHERE battle_id = ${id}
+      ORDER BY created_at ASC
+    `;
+
+    return { battle, contestants, cross_examinations: crossExaminations };
+  });
+
+  // POST /api/battles/:id/stop — cancel a running battle
+  app.post<{ Params: { id: string } }>('/api/battles/:id/stop', async (req, reply) => {
+    const parsedId = UuidParam.safeParse(req.params.id);
+    if (!parsedId.success) {
+      reply.code(400);
+      return { error: 'invalid id' };
+    }
+    const id = parsedId.data;
+
+    const [row] = await sql<{ id: string; status: string }[]>`
+      SELECT id, status FROM battles WHERE id = ${id}
+    `;
+    if (!row) {
+      reply.code(404);
+      return { error: 'battle not found' };
+    }
+    if (row.status !== 'running') {
+      reply.code(409);
+      return { error: `cannot stop battle in status '${row.status}'` };
+    }
+
+    const { cancelled, taskIds } = await battleRunner.cancelBattle(id);
+    if (!cancelled) {
+      reply.code(409);
+      return { error: 'battle is no longer running' };
+    }
+
+    // Abort any in-flight dispatcher tasks (cloud contestants running externally).
+    for (const taskId of taskIds) {
+      cancelExternal(taskId);
+    }
+
+    return { cancelled: true };
+  });
+
+  // GET /api/battles/:id/analysis — read analysis.md from the battle's results_path
+  app.get<{ Params: { id: string } }>('/api/battles/:id/analysis', async (req, reply) => {
+    const parsedId = UuidParam.safeParse(req.params.id);
+    if (!parsedId.success) {
+      reply.code(400);
+      return { error: 'invalid id' };
+    }
+    const id = parsedId.data;
+
+    const [row] = await sql<{ results_path: string | null }[]>`
+      SELECT results_path FROM battles WHERE id = ${id}
+    `;
+    if (!row) {
+      reply.code(404);
+      return { error: 'battle not found' };
+    }
+    if (!row.results_path) {
+      reply.code(404);
+      return { error: 'analysis not ready' };
+    }
+
+    try {
+      const text = await readFile(join(row.results_path, 'analysis.md'), 'utf8');
+      return { text };
+    } catch {
+      reply.code(404);
+      return { error: 'analysis not ready' };
+    }
+  });
+
+  // POST /api/battles/:id/analyze — trigger or re-trigger analysis
+  app.post<{ Params: { id: string } }>('/api/battles/:id/analyze', async (req, reply) => {
+    const parsedId = UuidParam.safeParse(req.params.id);
+    if (!parsedId.success) {
+      reply.code(400);
+      return { error: 'invalid id' };
+    }
+    const id = parsedId.data;
+
+    const [row] = await sql<{ id: string; status: string }[]>`
+      SELECT id, status FROM battles WHERE id = ${id}
+    `;
+    if (!row) {
+      reply.code(404);
+      return { error: 'battle not found' };
+    }
+    if (row.status === 'running') {
+      reply.code(409);
+      return { error: 'battle is still running — wait for all contestants to finish' };
+    }
+
+    const result = await battleRunner.triggerAnalysis(id);
+    if (!result.triggered) {
+      reply.code(404);
+      return { error: 'battle not found' };
+    }
+
+    reply.code(202);
+    return { triggered: true };
+  });
+
+  // PATCH /api/battles/:id/winner — manually set or clear the winner.
+  // Validates the contestant belongs to the battle; publishes battle_updated so
+  // the pane badge reflects the override immediately. Human is authoritative.
+  app.patch<{ Params: { id: string } }>('/api/battles/:id/winner', async (req, reply) => {
+    const parsedId = UuidParam.safeParse(req.params.id);
+    if (!parsedId.success) {
+      reply.code(400);
+      return { error: 'invalid id' };
+    }
+
+    const parsed = SetWinnerBody.safeParse(req.body);
+    if (!parsed.success) {
+      reply.code(400);
+      return { error: 'invalid body', details: parsed.error.flatten() };
+    }
+
+    const result = await battleRunner.setWinner(parsedId.data, parsed.data.winner_contestant_id);
+    if (!result.ok) {
+      if (result.notFound) { reply.code(404); return { error: 'battle not found' }; }
+      if (result.invalidContestant) { reply.code(422); return { error: 'contestant not found in this battle' }; }
+      reply.code(500); return { error: 'unknown error' };
+    }
+    return { ok: true };
+  });
+
+  // GET /api/battles/:id/contestants/:cid/diff — read the diff.patch for a coding contestant.
+  app.get<{ Params: { id: string; cid: string } }>('/api/battles/:id/contestants/:cid/diff', async (req, reply) => {
+    const parsedId = UuidParam.safeParse(req.params.id);
+    const parsedCid = UuidParam.safeParse(req.params.cid);
+    if (!parsedId.success || !parsedCid.success) {
+      reply.code(400);
+      return { error: 'invalid id' };
+    }
+
+    const [contestant] = await sql<{ result_path: string | null }[]>`
+      SELECT result_path FROM contestants
+      WHERE id = ${parsedCid.data} AND battle_id = ${parsedId.data}
+    `;
+    if (!contestant) {
+      reply.code(404);
+      return { error: 'contestant not found' };
+    }
+    if (!contestant.result_path) {
+      reply.code(404);
+      return { error: 'diff not available' };
+    }
+
+    try {
+      const text = await readFile(join(contestant.result_path, 'diff.patch'), 'utf8');
+      return { diff: text };
+    } catch {
+      reply.code(404);
+      return { error: 'diff not available' };
+    }
+  });
+
+  // POST /api/battles/:id/cross-examine — start a cross-examination
+  app.post<{ Params: { id: string } }>('/api/battles/:id/cross-examine', async (req, reply) => {
+    const parsedId = UuidParam.safeParse(req.params.id);
+    if (!parsedId.success) {
+      reply.code(400);
+      return { error: 'invalid id' };
+    }
+    const id = parsedId.data;
+
+    const parsed = CrossExamineBody.safeParse(req.body);
+    if (!parsed.success) {
+      reply.code(400);
+      return { error: 'invalid body', details: parsed.error.flatten() };
+    }
+
+    const [row] = await sql<{ id: string; status: string }[]>`
+      SELECT id, status FROM battles WHERE id = ${id}
+    `;
+    if (!row) {
+      reply.code(404);
+      return { error: 'battle not found' };
+    }
+    if (row.status === 'running') {
+      reply.code(409);
+      return { error: 'battle is still running — cross-examine after all contestants finish' };
+    }
+
+    const { crossExamId } = await battleRunner.startCrossExam(id, {
+      identity: parsed.data.identity,
+      model: parsed.data.model,
+    });
+
+    reply.code(202);
+    return { cross_exam_id: crossExamId };
+  });
 }

apps/coder/src/routes/messages.ts

@@ -4,6 +4,7 @@ import type { Sql } from '../db.js';
 import type { Broker } from '@boocode/server/broker';
 import type { WsFrame } from '@boocode/contracts/ws-frames';
 import { resolveChatId } from './chat-resolve.js';
+import { asPermissionMode } from '../services/tools/types.js';
 
 const AnswerUserInputBody = z.object({
   tool_call_id: z.string().min(1),
@@ -43,7 +44,13 @@ const SendBody = z.object({
 });
 
 interface InferenceApi {
-  enqueue: (sessionId: string, chatId: string, assistantId: string, user: string) => void;
+  enqueue: (
+    sessionId: string,
+    chatId: string,
+    assistantId: string,
+    user: string,
+    permissionMode?: 'plan' | 'ask' | 'bypass',
+  ) => void;
   cancel: (sessionId: string, chatId: string) => Promise<boolean>;
   hasActive: (chatId: string) => boolean;
 }
@@ -245,7 +252,16 @@ export function registerMessageRoutes(
         RETURNING id
       `;
 
-      inference.enqueue(sessionId, chatId, assistantMsg!.id, 'default');
+      // Native BooCode permission gate (plan/ask/bypass) — threaded into the
+      // write-tool context so create/edit/delete and apply_pending honor it.
+      // Plan = read-only, Ask = stage to the queue (agent can't self-apply),
+      // Bypass = apply each write immediately. Other mode ids (e.g. an external
+      // fallback's native mode) leave the gate undefined = legacy behavior.
+      req.log.info(
+        { provider, mode_id, permissionMode: asPermissionMode(mode_id), chatId },
+        'native enqueue — permission gate',
+      );
+      inference.enqueue(sessionId, chatId, assistantMsg!.id, 'default', asPermissionMode(mode_id));
 
       reply.code(202);
       return { user_message_id: userMsg!.id, assistant_message_id: assistantMsg!.id };

apps/coder/src/schema.sql

@@ -54,9 +54,6 @@ DO $$ BEGIN
   END IF;
 END $$;
 
--- v2.0.5: arena support — group tasks into competitive arenas.
-ALTER TABLE tasks ADD COLUMN IF NOT EXISTS arena_id UUID;
-
 -- Human inbox: tasks needing attention
 CREATE OR REPLACE VIEW human_inbox AS
   SELECT * FROM tasks WHERE state IN ('blocked', 'failed');
@@ -81,6 +78,7 @@ ALTER TABLE tasks ADD COLUMN IF NOT EXISTS thinking_option_id TEXT;
 DROP VIEW IF EXISTS human_inbox;
 ALTER TABLE tasks DROP COLUMN IF EXISTS feature_values;
 ALTER TABLE tasks DROP COLUMN IF EXISTS worktree_path;
+ALTER TABLE tasks DROP COLUMN IF EXISTS arena_id;
 CREATE OR REPLACE VIEW human_inbox AS
   SELECT * FROM tasks WHERE state IN ('blocked', 'failed');
 
@@ -157,7 +155,7 @@ CREATE UNIQUE INDEX IF NOT EXISTS worktrees_active_path_uidx ON worktrees(path)
 DROP TABLE IF EXISTS session_worktrees;
 
 -- Dispatch hint: which chat (tab) a task belongs to. The coder message route and
--- skills route set it from the frontend tab; session-less creators (arena, MCP,
+-- skills route set it from the frontend tab; session-less creators (MCP,
 -- new_task, generic /api/tasks) leave it NULL and the dispatcher creates a chat.
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS chat_id UUID REFERENCES chats(id) ON DELETE SET NULL;
 
@@ -271,7 +269,7 @@ ALTER TABLE agent_sessions ADD CONSTRAINT agent_sessions_backend_chk
   CHECK (backend IN ('opencode_server', 'acp_warm', 'claude_sdk'));
 
 -- LISTEN/NOTIFY fast path: every tasks INSERT (from any call site — routes,
--- new_task tool, arena, MCP server) fires pg_notify('tasks_new') in the same
+-- new_task tool, MCP server) fires pg_notify('tasks_new') in the same
 -- transaction, so the dispatcher reacts immediately instead of waiting for the
 -- fallback poll. Postgres holds the notification until COMMIT, so the listener
 -- always sees the committed row. A trigger covers all insert paths with no
@@ -357,3 +355,86 @@ DO $$ BEGIN
       CHECK (status IN ('pending', 'running', 'completed', 'failed', 'skipped', 'cancelled'));
   END IF;
 END $$;
+
+-- Arena: battles + contestants + cross_examinations.
+-- project_id carries no FK (matches tasks.project_id + flow_runs.project_id convention).
+-- winner_contestant_id FK is deferred (forward reference): added via guarded ALTER below.
+CREATE TABLE IF NOT EXISTS battles (
+  id                   UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+  project_id           UUID NOT NULL,
+  battle_type          TEXT NOT NULL,
+  prompt               TEXT NOT NULL,
+  status               TEXT NOT NULL DEFAULT 'pending',
+  winner_contestant_id UUID,
+  results_path         TEXT,
+  error                TEXT,
+  created_at           TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp(),
+  updated_at           TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp(),
+  CONSTRAINT battles_type_chk   CHECK (battle_type IN ('coding', 'qa')),
+  CONSTRAINT battles_status_chk CHECK (status IN ('pending', 'running', 'completed', 'failed', 'cancelled'))
+);
+
+CREATE TABLE IF NOT EXISTS contestants (
+  id             UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+  battle_id      UUID NOT NULL REFERENCES battles(id) ON DELETE CASCADE,
+  identity       TEXT NOT NULL,
+  model          TEXT NOT NULL,
+  lane           TEXT NOT NULL,
+  task_id        UUID REFERENCES tasks(id) ON DELETE SET NULL,
+  worktree_id    UUID REFERENCES worktrees(id) ON DELETE SET NULL,
+  status         TEXT NOT NULL DEFAULT 'queued',
+  duration_ms    INTEGER,
+  tokens_per_sec DOUBLE PRECISION,
+  cost_tokens    INTEGER,
+  result_path    TEXT,
+  error          TEXT,
+  created_at     TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp(),
+  updated_at     TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp(),
+  CONSTRAINT contestants_lane_chk   CHECK (lane   IN ('local', 'cloud')),
+  CONSTRAINT contestants_status_chk CHECK (status IN ('queued', 'running', 'done', 'error')),
+  UNIQUE (battle_id, identity, model)
+);
+
+CREATE TABLE IF NOT EXISTS cross_examinations (
+  id         UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+  battle_id  UUID NOT NULL REFERENCES battles(id) ON DELETE CASCADE,
+  identity   TEXT NOT NULL,
+  model      TEXT NOT NULL,
+  verdict    TEXT,
+  created_at TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp()
+);
+
+-- Add the winner FK now that contestants exists.
+DO $$ BEGIN
+  IF NOT EXISTS (SELECT 1 FROM pg_constraint WHERE conname = 'battles_winner_contestant_id_fkey') THEN
+    ALTER TABLE battles ADD CONSTRAINT battles_winner_contestant_id_fkey
+      FOREIGN KEY (winner_contestant_id) REFERENCES contestants(id) ON DELETE SET NULL;
+  END IF;
+END $$;
+
+-- battles query (GET /api/battles?project_id=).
+CREATE INDEX IF NOT EXISTS battles_project_created_idx ON battles(project_id, created_at DESC);
+
+-- Lane-scheduler advance scans (contestants WHERE battle_id = ? AND status = ?).
+CREATE INDEX IF NOT EXISTS contestants_battle_status_idx ON contestants(battle_id, status);
+
+-- onTaskTerminal callback: look up the contestant owning a completed task.
+CREATE INDEX IF NOT EXISTS contestants_task_id_idx ON contestants(task_id);
+
+-- Cross-examination listing per battle.
+CREATE INDEX IF NOT EXISTS cross_examinations_battle_idx ON cross_examinations(battle_id);
+
+-- TokenScope: per-category token breakdown on arena contestants and tasks.
+ALTER TABLE contestants ADD COLUMN IF NOT EXISTS token_breakdown JSONB;
+ALTER TABLE tasks ADD COLUMN IF NOT EXISTS token_breakdown JSONB;
+
+-- Orchestrator flow step events (append-only event log for resume/replay).
+CREATE TABLE IF NOT EXISTS flow_step_events (
+  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+  run_id UUID NOT NULL REFERENCES flow_runs(id),
+  step_id VARCHAR(64) NOT NULL,
+  event VARCHAR(32) NOT NULL,
+  payload JSONB,
+  created_at TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp()
+);
+CREATE INDEX IF NOT EXISTS flow_step_events_run_idx ON flow_step_events(run_id);

apps/coder/src/services/__tests__/arena-analyzer-helpers.test.ts

@@ -0,0 +1,254 @@
+import { describe, it, expect } from 'vitest';
+import {
+  buildDigestPrompt,
+  buildJudgePrompt,
+  buildCrossExamPrompt,
+  extractWinner,
+  shouldNameWinner,
+  type ContestantDigest,
+  type ContestantDigestInput,
+} from '../arena-analyzer-helpers.js';
+
+// ─── shouldNameWinner ─────────────────────────────────────────────────────────
+
+describe('shouldNameWinner', () => {
+  it('returns false with 0 succeeded contestants', () => {
+    expect(shouldNameWinner(0)).toBe(false);
+  });
+
+  it('returns false with exactly 1 succeeded contestant', () => {
+    expect(shouldNameWinner(1)).toBe(false);
+  });
+
+  it('returns true with exactly 2 succeeded contestants', () => {
+    expect(shouldNameWinner(2)).toBe(true);
+  });
+
+  it('returns true with more than 2 succeeded contestants', () => {
+    expect(shouldNameWinner(3)).toBe(true);
+    expect(shouldNameWinner(6)).toBe(true);
+  });
+});
+
+// ─── extractWinner ────────────────────────────────────────────────────────────
+
+describe('extractWinner', () => {
+  it('extracts identity and model from a WINNER: line', () => {
+    const output = 'Some analysis\n\nWINNER: claude/opus-4-5\n\nMore text.';
+    expect(extractWinner(output)).toEqual({ identity: 'claude', model: 'opus-4-5' });
+  });
+
+  it('is case-insensitive for the WINNER keyword', () => {
+    expect(extractWinner('winner: boocode/qwen3.6-35b')).toEqual({
+      identity: 'boocode',
+      model: 'qwen3.6-35b',
+    });
+    expect(extractWinner('Winner: opencode/some-model')).toEqual({
+      identity: 'opencode',
+      model: 'some-model',
+    });
+  });
+
+  it('returns null when NO_WINNER is declared', () => {
+    expect(extractWinner('WINNER: NO_WINNER')).toBeNull();
+    expect(extractWinner('winner: no_winner')).toBeNull();
+  });
+
+  it('returns null when no WINNER line is present', () => {
+    expect(extractWinner('Just some analysis text with no verdict.')).toBeNull();
+    expect(extractWinner('')).toBeNull();
+  });
+
+  it('returns null when the WINNER line has no slash separator', () => {
+    expect(extractWinner('WINNER: justidentity')).toBeNull();
+  });
+
+  it('returns null when the WINNER line is empty after the colon', () => {
+    expect(extractWinner('WINNER:')).toBeNull();
+    expect(extractWinner('WINNER:   ')).toBeNull();
+  });
+
+  it('handles leading and trailing whitespace around the slash parts', () => {
+    const result = extractWinner('WINNER:  claude / opus-4-5 ');
+    expect(result).toEqual({ identity: 'claude', model: 'opus-4-5' });
+  });
+
+  it('picks the first WINNER line when multiple are present', () => {
+    const output = 'WINNER: claude/opus-4-5\nWINNER: opencode/other-model';
+    expect(extractWinner(output)).toEqual({ identity: 'claude', model: 'opus-4-5' });
+  });
+
+  it('handles model names that contain slashes by splitting at the first slash only', () => {
+    // edge case: model name with a slash — should still split at first slash
+    // identity = 'native', model = 'llama-swap/qwen3.6'
+    const result = extractWinner('WINNER: native/llama-swap/qwen3.6');
+    expect(result).toEqual({ identity: 'native', model: 'llama-swap/qwen3.6' });
+  });
+});
+
+// ─── buildDigestPrompt ────────────────────────────────────────────────────────
+
+describe('buildDigestPrompt', () => {
+  const base: ContestantDigestInput = {
+    identity: 'claude',
+    model: 'opus-4-5',
+    resultMd: '# Output\n\nSome result content.',
+    benchmarkLine: '12000ms',
+  };
+
+  it('returns an object with non-empty system and user strings', () => {
+    const { system, user } = buildDigestPrompt(base);
+    expect(system.length).toBeGreaterThan(0);
+    expect(user.length).toBeGreaterThan(0);
+  });
+
+  it('includes the contestant identity and model in the user prompt', () => {
+    const { user } = buildDigestPrompt(base);
+    expect(user).toContain('claude');
+    expect(user).toContain('opus-4-5');
+  });
+
+  it('includes the benchmark line in the user prompt', () => {
+    const { user } = buildDigestPrompt(base);
+    expect(user).toContain('12000ms');
+  });
+
+  it('includes the result.md content in the user prompt', () => {
+    const { user } = buildDigestPrompt(base);
+    expect(user).toContain('Some result content.');
+  });
+
+  it('includes the diff.patch when provided', () => {
+    const input: ContestantDigestInput = { ...base, diffPatch: '--- a/foo.ts\n+++ b/foo.ts\n+added' };
+    const { user } = buildDigestPrompt(input);
+    expect(user).toContain('added');
+    expect(user).toContain('```diff');
+  });
+
+  it('omits the diff section when diffPatch is undefined', () => {
+    const { user } = buildDigestPrompt(base);
+    expect(user).not.toContain('```diff');
+  });
+
+  it('truncates resultMd longer than 8000 characters', () => {
+    const longResult = 'x'.repeat(10_000);
+    const { user } = buildDigestPrompt({ ...base, resultMd: longResult });
+    // The truncated content must not exceed 8000 chars in the sliced section.
+    // We just check the total user string doesn't balloon unreasonably.
+    expect(user.length).toBeLessThan(15_000);
+  });
+
+  it('truncates diffPatch longer than 5000 characters', () => {
+    const longDiff = '+' + 'x'.repeat(10_000);
+    const { user } = buildDigestPrompt({ ...base, diffPatch: longDiff });
+    expect(user.length).toBeLessThan(16_000);
+  });
+});
+
+// ─── buildJudgePrompt ─────────────────────────────────────────────────────────
+
+describe('buildJudgePrompt', () => {
+  const digests: ContestantDigest[] = [
+    { identity: 'claude', model: 'opus-4-5', digest: 'Good result.', benchmarkLine: '5000ms' },
+    { identity: 'opencode', model: 'qwen3.6', digest: 'Decent result.', benchmarkLine: '8000ms' },
+  ];
+
+  it('includes the original prompt in the user section', () => {
+    const { user } = buildJudgePrompt('Write a sorting algorithm', digests);
+    expect(user).toContain('Write a sorting algorithm');
+  });
+
+  it('includes each contestant heading in the user section', () => {
+    const { user } = buildJudgePrompt('prompt', digests);
+    expect(user).toContain('claude');
+    expect(user).toContain('opus-4-5');
+    expect(user).toContain('opencode');
+    expect(user).toContain('qwen3.6');
+  });
+
+  it('includes each contestant digest text', () => {
+    const { user } = buildJudgePrompt('prompt', digests);
+    expect(user).toContain('Good result.');
+    expect(user).toContain('Decent result.');
+  });
+
+  it('instructs the model to name a WINNER when 2+ digests are provided', () => {
+    const { system } = buildJudgePrompt('prompt', digests);
+    expect(system).toContain('WINNER:');
+  });
+
+  it('instructs the model NOT to name a winner when fewer than 2 digests are provided', () => {
+    const oneDigest = digests.slice(0, 1);
+    const { system } = buildJudgePrompt('prompt', oneDigest);
+    expect(system).toContain('NO_WINNER');
+    expect(system).not.toContain('WINNER: <identity>');
+  });
+
+  it('instructs NO_WINNER when digests list is empty', () => {
+    const { system } = buildJudgePrompt('prompt', []);
+    expect(system).toContain('NO_WINNER');
+  });
+
+  it('truncates originalPrompt longer than 2000 characters', () => {
+    const longPrompt = 'p'.repeat(5_000);
+    const { user } = buildJudgePrompt(longPrompt, digests);
+    // Should not contain more than 2000 chars of the prompt.
+    const promptSection = user.split('# Contestant Digests')[0] ?? '';
+    expect(promptSection.length).toBeLessThan(3_000);
+  });
+});
+
+// ─── buildCrossExamPrompt ─────────────────────────────────────────────────────
+
+describe('buildCrossExamPrompt', () => {
+  const digests: ContestantDigest[] = [
+    { identity: 'claude', model: 'opus-4-5', digest: 'Strong result.', benchmarkLine: '5000ms' },
+    { identity: 'boocode', model: 'qwen3.6-35b', digest: 'Decent result.', benchmarkLine: '12000ms' },
+  ];
+
+  const baseOpts = {
+    originalPrompt: 'Write a sorting algorithm.',
+    digests,
+    analysisContent: '# Arena Analysis\n\nClaude did better.\n\nWINNER: claude/opus-4-5',
+    proposedWinner: 'claude/opus-4-5',
+    examinerIdentity: 'goose',
+    examinerModel: 'gpt-4o',
+  };
+
+  it('includes the examiner identity and model in the system prompt', () => {
+    const { system } = buildCrossExamPrompt(baseOpts);
+    expect(system).toContain('goose');
+    expect(system).toContain('gpt-4o');
+  });
+
+  it('includes the original prompt in the user section', () => {
+    const { user } = buildCrossExamPrompt(baseOpts);
+    expect(user).toContain('Write a sorting algorithm.');
+  });
+
+  it('includes each contestant digest', () => {
+    const { user } = buildCrossExamPrompt(baseOpts);
+    expect(user).toContain('Strong result.');
+    expect(user).toContain('Decent result.');
+  });
+
+  it('includes the proposed analysis content', () => {
+    const { user } = buildCrossExamPrompt(baseOpts);
+    expect(user).toContain('Claude did better.');
+  });
+
+  it('includes the proposed winner when set', () => {
+    const { user } = buildCrossExamPrompt(baseOpts);
+    expect(user).toContain('claude/opus-4-5');
+  });
+
+  it('notes that no winner was proposed when proposedWinner is null', () => {
+    const { user } = buildCrossExamPrompt({ ...baseOpts, proposedWinner: null });
+    expect(user).toContain('No winner was proposed');
+  });
+
+  it('instructs the examiner to provide a VERDICT line', () => {
+    const { system } = buildCrossExamPrompt(baseOpts);
+    expect(system).toContain('VERDICT:');
+  });
+});

apps/coder/src/services/__tests__/arena-decisions.test.ts

@@ -0,0 +1,350 @@
+import { describe, it, expect } from 'vitest';
+import {
+  classifyLane,
+  nextLocalContestant,
+  isBattleComplete,
+  computeBenchmark,
+  sanitizeSlug,
+  buildBattleSlug,
+  buildContestantDir,
+  reconcileContestantResume,
+  reconcileContestants,
+  type ContestantSlot,
+} from '../arena-decisions.js';
+
+// Local models = what the llama-swap server actually serves.
+const LOCAL_MODELS: ReadonlySet<string> = new Set([
+  'qwen3.6-35b-a3b-mxfp4',
+  'qwen2.5-coder-7b',
+]);
+
+// ─── classifyLane ────────────────────────────────────────────────────────────
+
+describe('classifyLane', () => {
+  it('classifies qa battles as local regardless of identity or model', () => {
+    expect(classifyLane('qa', 'boocode', 'qwen3.6-35b-a3b-mxfp4', LOCAL_MODELS)).toBe('local');
+    expect(classifyLane('qa', 'claude', 'claude-opus-4-5', LOCAL_MODELS)).toBe('local');
+    expect(classifyLane('qa', 'Debugger', 'cloud-model', new Set())).toBe('local');
+    expect(classifyLane('qa', 'opencode', 'any-model', LOCAL_MODELS)).toBe('local');
+  });
+
+  it('classifies coding contestants as local when model is in localModels', () => {
+    expect(classifyLane('coding', 'boocode', 'qwen3.6-35b-a3b-mxfp4', LOCAL_MODELS)).toBe('local');
+    expect(classifyLane('coding', 'opencode', 'qwen3.6-35b-a3b-mxfp4', LOCAL_MODELS)).toBe('local');
+    expect(classifyLane('coding', 'qwen', 'qwen2.5-coder-7b', LOCAL_MODELS)).toBe('local');
+  });
+
+  it('classifies coding contestants as cloud when model is not in localModels', () => {
+    expect(classifyLane('coding', 'claude', 'claude-opus-4-5', LOCAL_MODELS)).toBe('cloud');
+    expect(classifyLane('coding', 'opencode', 'claude-opus-4-5', LOCAL_MODELS)).toBe('cloud');
+    expect(classifyLane('coding', 'goose', 'gpt-4o', LOCAL_MODELS)).toBe('cloud');
+    expect(classifyLane('coding', 'qwen', 'unknown-remote-model', LOCAL_MODELS)).toBe('cloud');
+  });
+
+  it('uses the injected localModels set, not a hardcoded list', () => {
+    const custom = new Set(['my-local-model']);
+    expect(classifyLane('coding', 'any-agent', 'my-local-model', custom)).toBe('local');
+    expect(classifyLane('coding', 'boocode', 'other-model', custom)).toBe('cloud');
+  });
+
+  it('defaults to cloud for an empty localModels set', () => {
+    expect(classifyLane('coding', 'boocode', 'qwen3.6-35b-a3b-mxfp4', new Set())).toBe('cloud');
+    expect(classifyLane('coding', 'native', 'any-local-model', new Set())).toBe('cloud');
+  });
+});
+
+// ─── nextLocalContestant ─────────────────────────────────────────────────────
+
+describe('nextLocalContestant', () => {
+  it('returns null for an empty list', () => {
+    expect(nextLocalContestant([])).toBeNull();
+  });
+
+  it('returns null when no local contestants are queued', () => {
+    const slots: ContestantSlot[] = [
+      { id: 'c1', lane: 'local', status: 'running' },
+      { id: 'c2', lane: 'cloud', status: 'queued' },
+    ];
+    expect(nextLocalContestant(slots)).toBeNull();
+  });
+
+  it('returns the first queued local contestant in order', () => {
+    const slots: ContestantSlot[] = [
+      { id: 'c1', lane: 'local', status: 'done' },
+      { id: 'c2', lane: 'local', status: 'queued' },
+      { id: 'c3', lane: 'local', status: 'queued' },
+    ];
+    expect(nextLocalContestant(slots)).toBe('c2');
+  });
+
+  it('skips done/error local contestants and cloud contestants', () => {
+    const slots: ContestantSlot[] = [
+      { id: 'c1', lane: 'cloud', status: 'queued' },
+      { id: 'c2', lane: 'local', status: 'error' },
+      { id: 'c3', lane: 'local', status: 'queued' },
+    ];
+    expect(nextLocalContestant(slots)).toBe('c3');
+  });
+
+  it('returns null when all local contestants are done or error', () => {
+    const slots: ContestantSlot[] = [
+      { id: 'c1', lane: 'local', status: 'done' },
+      { id: 'c2', lane: 'local', status: 'error' },
+    ];
+    expect(nextLocalContestant(slots)).toBeNull();
+  });
+});
+
+// ─── isBattleComplete ────────────────────────────────────────────────────────
+
+describe('isBattleComplete', () => {
+  it('returns false for an empty list', () => {
+    expect(isBattleComplete([])).toBe(false);
+  });
+
+  it('returns true when all contestants are done', () => {
+    expect(isBattleComplete([{ status: 'done' }, { status: 'done' }])).toBe(true);
+  });
+
+  it('returns true when all contestants are error', () => {
+    expect(isBattleComplete([{ status: 'error' }, { status: 'error' }])).toBe(true);
+  });
+
+  it('returns true for a mixed done/error result', () => {
+    expect(isBattleComplete([{ status: 'done' }, { status: 'error' }, { status: 'done' }])).toBe(true);
+  });
+
+  it('returns false while any contestant is still running', () => {
+    expect(isBattleComplete([{ status: 'done' }, { status: 'running' }])).toBe(false);
+  });
+
+  it('returns false while any contestant is still queued', () => {
+    expect(isBattleComplete([{ status: 'done' }, { status: 'queued' }])).toBe(false);
+  });
+});
+
+// ─── computeBenchmark ────────────────────────────────────────────────────────
+
+describe('computeBenchmark', () => {
+  const t0 = new Date('2026-06-06T10:00:00.000Z');
+  const t1 = new Date('2026-06-06T10:00:05.000Z'); // +5 000ms
+
+  it('computes duration in ms for both lanes', () => {
+    const local = computeBenchmark(t0, t1, 100, 'local');
+    expect(local.durationMs).toBe(5000);
+    const cloud = computeBenchmark(t0, t1, null, 'cloud');
+    expect(cloud.durationMs).toBe(5000);
+  });
+
+  it('computes tokens/sec for local lane when costTokens is known', () => {
+    const bench = computeBenchmark(t0, t1, 500, 'local');
+    expect(bench.tokensPerSec).toBeCloseTo(100, 5); // 500 / 5 = 100 tok/s
+  });
+
+  it('omits tokens/sec for cloud lane regardless of costTokens', () => {
+    const bench = computeBenchmark(t0, t1, 500, 'cloud');
+    expect(bench.tokensPerSec).toBeNull();
+  });
+
+  it('omits tokens/sec for local lane when costTokens is null', () => {
+    const bench = computeBenchmark(t0, t1, null, 'local');
+    expect(bench.tokensPerSec).toBeNull();
+  });
+
+  it('returns durationMs = 0 and null tokensPerSec when timestamps are equal', () => {
+    const bench = computeBenchmark(t0, t0, 100, 'local');
+    expect(bench.durationMs).toBe(0);
+    expect(bench.tokensPerSec).toBeNull();
+  });
+
+  it('clamps negative duration to 0 (clock skew)', () => {
+    const bench = computeBenchmark(t1, t0, 50, 'local');
+    expect(bench.durationMs).toBe(0);
+    expect(bench.tokensPerSec).toBeNull();
+  });
+
+  it('includes token breakdown when provided', () => {
+    const breakdown = {
+      system: 10,
+      user: 20,
+      assistant: 30,
+      tools: 40,
+      reasoning: 5,
+      total: 105,
+    };
+    const bench = computeBenchmark(t0, t1, 500, 'local', breakdown);
+    expect(bench.tokenBreakdown).toEqual(breakdown);
+  });
+
+  it('defaults token breakdown to null when omitted', () => {
+    const bench = computeBenchmark(t0, t1, 500, 'local');
+    expect(bench.tokenBreakdown).toBeNull();
+  });
+});
+
+// ─── sanitizeSlug ────────────────────────────────────────────────────────────
+
+describe('sanitizeSlug', () => {
+  it('lowercases and preserves alphanumeric + hyphens', () => {
+    expect(sanitizeSlug('claude')).toBe('claude');
+    expect(sanitizeSlug('claude-opus-4-5')).toBe('claude-opus-4-5');
+  });
+
+  it('replaces spaces and special characters with hyphens', () => {
+    expect(sanitizeSlug('Code Reviewer')).toBe('code-reviewer');
+    expect(sanitizeSlug('native/boocode')).toBe('native-boocode');
+    expect(sanitizeSlug('qwen2.5-coder-35b')).toBe('qwen2-5-coder-35b');
+  });
+
+  it('collapses consecutive non-alphanumeric runs to a single hyphen', () => {
+    expect(sanitizeSlug('foo  bar---baz')).toBe('foo-bar-baz');
+  });
+
+  it('strips leading and trailing hyphens', () => {
+    expect(sanitizeSlug('---foo---')).toBe('foo');
+  });
+
+  it('truncates to 64 characters', () => {
+    const long = 'a'.repeat(100);
+    expect(sanitizeSlug(long).length).toBe(64);
+  });
+});
+
+// ─── buildBattleSlug ─────────────────────────────────────────────────────────
+
+describe('buildBattleSlug', () => {
+  it('builds a deterministic dated slug from id, type, and createdAt', () => {
+    const id = 'a1b2c3d4-e5f6-7890-abcd-ef1234567890';
+    const createdAt = new Date('2026-06-06T12:00:00.000Z');
+    const slug = buildBattleSlug(id, 'coding', createdAt);
+    expect(slug).toBe('2026-06-06-coding-a1b2c3d4');
+  });
+
+  it('includes the battle type in the slug', () => {
+    const id = 'aaaaaaaa-0000-0000-0000-000000000000';
+    const createdAt = new Date('2026-01-01T00:00:00.000Z');
+    expect(buildBattleSlug(id, 'qa', createdAt)).toContain('-qa-');
+    expect(buildBattleSlug(id, 'coding', createdAt)).toContain('-coding-');
+  });
+
+  it('uses the first 8 hex chars of the uuid (dashes stripped)', () => {
+    const id = 'deadbeef-0000-0000-0000-000000000000';
+    const slug = buildBattleSlug(id, 'coding', new Date('2026-06-06T00:00:00Z'));
+    expect(slug.endsWith('-deadbeef')).toBe(true);
+  });
+});
+
+// ─── buildContestantDir ──────────────────────────────────────────────────────
+
+describe('buildContestantDir', () => {
+  it('joins sanitized identity and model with a hyphen', () => {
+    expect(buildContestantDir('claude', 'claude-opus-4-5')).toBe('claude-claude-opus-4-5');
+  });
+
+  it('sanitizes both parts independently', () => {
+    expect(buildContestantDir('Code Reviewer', 'qwen2.5-35b')).toBe('code-reviewer-qwen2-5-35b');
+  });
+});
+
+// ─── reconcileContestantResume ───────────────────────────────────────────────
+
+describe('reconcileContestantResume', () => {
+  it('keeps non-running contestants regardless of task state', () => {
+    for (const status of ['queued', 'done', 'error']) {
+      expect(reconcileContestantResume(status, 'tid', 'completed')).toBe('keep');
+      expect(reconcileContestantResume(status, null, null)).toBe('keep');
+    }
+  });
+
+  it('re-dispatches a running contestant with no task_id', () => {
+    expect(reconcileContestantResume('running', null, null)).toBe('re-dispatch');
+  });
+
+  it('re-dispatches a running contestant whose task row is absent', () => {
+    expect(reconcileContestantResume('running', 'tid', null)).toBe('re-dispatch');
+  });
+
+  it('marks done when the task completed before the terminal callback ran', () => {
+    expect(reconcileContestantResume('running', 'tid', 'completed')).toBe('mark-done');
+  });
+
+  it('marks error when the task failed', () => {
+    expect(reconcileContestantResume('running', 'tid', 'failed')).toBe('mark-error');
+  });
+
+  it('marks cancelled when the task was cancelled', () => {
+    expect(reconcileContestantResume('running', 'tid', 'cancelled')).toBe('mark-cancelled');
+  });
+
+  it('keeps a running contestant whose task is pending (dispatcher handles it)', () => {
+    expect(reconcileContestantResume('running', 'tid', 'pending')).toBe('keep');
+  });
+
+  it('re-dispatches when the task is stuck running (process died)', () => {
+    expect(reconcileContestantResume('running', 'tid', 'running')).toBe('re-dispatch');
+  });
+
+  it('re-dispatches when the task is blocked (permission dialog gone on restart)', () => {
+    expect(reconcileContestantResume('running', 'tid', 'blocked')).toBe('re-dispatch');
+  });
+});
+
+// ─── reconcileContestants ────────────────────────────────────────────────────
+
+describe('reconcileContestants', () => {
+  it('returns one decision per contestant', () => {
+    const contestants = [
+      { contestantId: 'c1', taskId: null, status: 'done' },
+      { contestantId: 'c2', taskId: 't1', status: 'running' },
+      { contestantId: 'c3', taskId: 't2', status: 'running' },
+    ];
+    const taskStates = new Map([['t1', 'completed'], ['t2', 'running']]);
+    const decisions = reconcileContestants(contestants, taskStates);
+    expect(decisions).toHaveLength(3);
+    expect(decisions[0]).toEqual({ contestantId: 'c1', action: 'keep' });
+    expect(decisions[1]).toEqual({ contestantId: 'c2', action: 'mark-done' });
+    expect(decisions[2]).toEqual({ contestantId: 'c3', action: 're-dispatch' });
+  });
+
+  it('re-dispatches a running contestant whose taskId is absent from taskStates', () => {
+    const contestants = [{ contestantId: 'c1', taskId: 'orphan', status: 'running' }];
+    const decisions = reconcileContestants(contestants, new Map());
+    expect(decisions[0]?.action).toBe('re-dispatch');
+  });
+
+  it('re-dispatches a running contestant with null taskId', () => {
+    const contestants = [{ contestantId: 'c1', taskId: null, status: 'running' }];
+    const decisions = reconcileContestants(contestants, new Map());
+    expect(decisions[0]?.action).toBe('re-dispatch');
+  });
+
+  it('returns empty array for no contestants', () => {
+    expect(reconcileContestants([], new Map())).toEqual([]);
+  });
+
+  it('keeps a running contestant whose task is pending', () => {
+    const contestants = [{ contestantId: 'c1', taskId: 't1', status: 'running' }];
+    const taskStates = new Map([['t1', 'pending']]);
+    const decisions = reconcileContestants(contestants, taskStates);
+    expect(decisions[0]?.action).toBe('keep');
+  });
+
+  it('handles a mixed battle: done/queued kept, stale running re-dispatched', () => {
+    const contestants = [
+      { contestantId: 'c1', taskId: 't1', status: 'done' },
+      { contestantId: 'c2', taskId: null, status: 'queued' },
+      { contestantId: 'c3', taskId: 't2', status: 'running' },
+      { contestantId: 'c4', taskId: 't3', status: 'running' },
+    ];
+    const taskStates = new Map([
+      ['t1', 'completed'],
+      ['t2', 'running'],  // stuck — process dead
+      ['t3', 'pending'],  // dispatcher will handle
+    ]);
+    const decisions = reconcileContestants(contestants, taskStates);
+    expect(decisions.find((d) => d.contestantId === 'c1')?.action).toBe('keep');
+    expect(decisions.find((d) => d.contestantId === 'c2')?.action).toBe('keep');
+    expect(decisions.find((d) => d.contestantId === 'c3')?.action).toBe('re-dispatch');
+    expect(decisions.find((d) => d.contestantId === 'c4')?.action).toBe('keep');
+  });
+});

apps/coder/src/services/__tests__/fuzzy-match.test.ts

@@ -161,6 +161,52 @@ describe('locateMatch — strategy 4: Levenshtein', () => {
   });
 });
 
+describe('locateMatch — strategy 4: fail-closed on ambiguity (corruption guard)', () => {
+  it('refuses (ambiguous) when two equally-similar anchored blocks both clear the bar', () => {
+    // The repetitive-file case that duplicated blocks: two blocks share the same
+    // first+last anchor lines and their middle lines are EQUALLY similar to the
+    // (drifted) needle. Tier 4 must refuse rather than splice over one of them.
+    const content = [
+      'const x = {',
+      '  total = aa;',
+      '};',
+      'const x = {',
+      '  total = bb;',
+      '};',
+    ].join('\n');
+    const needle = ['const x = {', '  total = ab;', '};'].join('\n');
+    const result = locateMatch(content, needle);
+    expect(result.kind).toBe('ambiguous');
+  });
+
+  it('refuses a below-threshold near-miss that the old 0.66 floor would have spliced', () => {
+    // ~0.7 similar: under the raised 0.85 floor this is now not_found, so the
+    // caller surfaces a correctable error instead of corrupting the file.
+    const content = 'const grandTotalAmount = a + b;\n';
+    const needle = 'const totalValue = a + b;';
+    const result = locateMatch(content, needle);
+    expect(result).toEqual({ kind: 'not_found' });
+  });
+
+  it('still matches a single genuine high-similarity drift uniquely', () => {
+    const content = 'const total = sum + tax;\n';
+    const needle = 'const totals = sum + tax;'; // one-char typo, ~0.96
+    const result = locateMatch(content, needle);
+    expect(result.kind).toBe('fuzzy');
+    const { start, end } = span(result);
+    expect(content.slice(start, end)).toBe('const total = sum + tax;');
+  });
+
+  it('requires an exact first+last line anchor for multi-line needles', () => {
+    // First line drifted too far to anchor → no window is scored → not_found,
+    // even though the middle lines are identical.
+    const content = ['function compute() {', '  return a + b;', '  return done;', '}'].join('\n');
+    const needle = ['totally different opener', '  return a + b;', '}'].join('\n');
+    const result = locateMatch(content, needle);
+    expect(result).toEqual({ kind: 'not_found' });
+  });
+});
+
 describe('locateMatch — edge cases', () => {
   it('returns not_found for an empty needle', () => {
     expect(locateMatch('anything', '')).toEqual({ kind: 'not_found' });

apps/coder/src/services/__tests__/pending_changes_integration.test.ts

@@ -83,6 +83,53 @@ describe.runIf(!!process.env.DATABASE_URL)('pending_changes integration', () =>
     expect(existsSync(resolve(testDir, 'deleteme.txt'))).toBe(false);
   });
 
+  it('re-emitted identical edits dedupe at queue and never duplicate on apply', async () => {
+    // Regression: the 2-3x block-stamping corruption. An anchored insert queued
+    // three times (a local model re-emitting the same tool call) must collapse to
+    // ONE pending row and apply exactly once.
+    await queueCreate(sql, testSessionId, null, 'dup.js', '<script>\nrender();\n', projectRoot)
+      .then((c) => applyOne(sql, c.id, projectRoot));
+
+    const oldStr = '<script>';
+    const newStr = '<script>\nconst recordFormats = ["gif"];';
+    const a = await queueEdit(sql, testSessionId, null, 'dup.js', oldStr, newStr, projectRoot);
+    const b = await queueEdit(sql, testSessionId, null, 'dup.js', oldStr, newStr, projectRoot);
+    const c = await queueEdit(sql, testSessionId, null, 'dup.js', oldStr, newStr, projectRoot);
+    // All three calls return the SAME pending row (deduped).
+    expect(b.id).toBe(a.id);
+    expect(c.id).toBe(a.id);
+
+    await applyOne(sql, a.id, projectRoot);
+    let content = await readFile(resolve(testDir, 'dup.js'), 'utf8');
+    expect((content.match(/const recordFormats/g) || []).length).toBe(1);
+
+    // Even a fresh, separately-queued identical edit re-applied is a no-op, not a stamp.
+    const again = await queueEdit(sql, testSessionId, null, 'dup.js', oldStr, newStr, projectRoot);
+    const res = await applyOne(sql, again.id, projectRoot);
+    expect(res.success).toBe(true);
+    content = await readFile(resolve(testDir, 'dup.js'), 'utf8');
+    expect((content.match(/const recordFormats/g) || []).length).toBe(1);
+  });
+
+  it('preserves CRLF line endings on edit', async () => {
+    await queueCreate(sql, testSessionId, null, 'crlf.txt', 'line one\r\nline two\r\nline three\r\n', projectRoot)
+      .then((c) => applyOne(sql, c.id, projectRoot));
+    const edit = await queueEdit(sql, testSessionId, null, 'crlf.txt', 'line two', 'line TWO', projectRoot);
+    const res = await applyOne(sql, edit.id, projectRoot);
+    expect(res.success).toBe(true);
+    const content = await readFile(resolve(testDir, 'crlf.txt'), 'utf8');
+    expect(content).toBe('line one\r\nline TWO\r\nline three\r\n');
+  });
+
+  it('refuses an edit that matches multiple locations instead of corrupting', async () => {
+    await queueCreate(sql, testSessionId, null, 'ambig.js', 'x=1;\ny=2;\nx=1;\n', projectRoot)
+      .then((ch) => applyOne(sql, ch.id, projectRoot));
+    const edit = await queueEdit(sql, testSessionId, null, 'ambig.js', 'x=1;', 'x=9;', projectRoot);
+    const res = await applyOne(sql, edit.id, projectRoot);
+    expect(res.success).toBe(false);
+    expect(res.error).toMatch(/matches 2 locations/);
+  });
+
   it('rewindOne → verify reverted', async () => {
     // Setup: create and apply a file
     const createChange = await queueCreate(sql, testSessionId, null, 'rewindable.txt', 'initial', projectRoot);

apps/coder/src/services/__tests__/plan-edit.test.ts

@@ -0,0 +1,69 @@
+import { describe, it, expect } from 'vitest';
+import { planEdit } from '../pending_changes.js';
+
+// planEdit is the pure core of applyOne's edit splice. These tests pin the
+// idempotency guards that stop the "block stamped 2-3x" corruption: applying the
+// same queued edit more than once must be a no-op, never a duplicate.
+
+describe('planEdit — normal application', () => {
+  it('applies a unique exact edit', () => {
+    const content = 'a\nfoo\nb\n';
+    const plan = planEdit(content, 'foo', 'bar');
+    expect(plan).toEqual({ kind: 'apply', updated: 'a\nbar\nb\n' });
+  });
+
+  it('reports ambiguous when old_string occurs more than once', () => {
+    const content = 'foo\nx\nfoo\n';
+    const plan = planEdit(content, 'foo', 'bar');
+    expect(plan).toEqual({ kind: 'ambiguous', count: 2 });
+  });
+
+  it('reports not_found when old_string is absent and new is not present', () => {
+    const content = 'alpha\nbeta\n';
+    const plan = planEdit(content, 'gamma that is clearly nowhere', 'delta');
+    expect(plan).toEqual({ kind: 'not_found' });
+  });
+});
+
+describe('planEdit — idempotency (the corruption guard)', () => {
+  it('treats a re-applied anchored insert as already-applied (no duplicate)', () => {
+    // The exact mechanism that tripled `const recordFormats` in settings.html:
+    // an anchored insert (old=anchor, new=anchor+block) where the anchor still
+    // matches uniquely after the first apply.
+    const oldStr = '<script>';
+    const newStr = '<script>\nconst recordFormats = ["gif","mp4"];';
+    const before = '<script>\nfunction render() {}\n</script>\n';
+
+    const first = planEdit(before, oldStr, newStr);
+    expect(first.kind).toBe('apply');
+    const after = first.kind === 'apply' ? first.updated : '';
+    expect((after.match(/const recordFormats/g) || []).length).toBe(1);
+
+    // Re-applying the identical edit to the already-edited content is a no-op.
+    const second = planEdit(after, oldStr, newStr);
+    expect(second).toEqual({ kind: 'noop', reason: 'already-applied' });
+  });
+
+  it('treats an edit whose old_string is gone but new_string is present as already-applied', () => {
+    const content = 'const total = sum + tax;\n';
+    const plan = planEdit(content, 'const subtotal = sum;', 'const total = sum + tax;');
+    expect(plan).toEqual({ kind: 'noop', reason: 'already-applied' });
+  });
+
+  it('treats a no-change splice as a noop', () => {
+    const content = 'a\nfoo\nb\n';
+    const plan = planEdit(content, 'foo', 'foo');
+    expect(plan).toEqual({ kind: 'noop', reason: 'identical' });
+  });
+
+  it('does not duplicate across three repeated applications', () => {
+    const oldStr = 'function f() {';
+    const newStr = 'function f() {\n  const x = 1;';
+    let content = 'function f() {\n  return x;\n}\n';
+    for (let i = 0; i < 3; i++) {
+      const plan = planEdit(content, oldStr, newStr);
+      if (plan.kind === 'apply') content = plan.updated;
+    }
+    expect((content.match(/const x = 1;/g) || []).length).toBe(1);
+  });
+});

apps/coder/src/services/__tests__/trigger-rules.test.ts

@@ -0,0 +1,31 @@
+import { describe, it, expect } from 'vitest';
+import { evaluateTriggerRule } from '../flow-runner-decisions.js';
+
+describe('evaluateTriggerRule', () => {
+  it('all_success requires all deps done', () => {
+    expect(evaluateTriggerRule(['a', 'b'], new Set(['a', 'b']), new Set(), new Set())).toBe(true);
+    expect(evaluateTriggerRule(['a', 'b'], new Set(['a']), new Set(), new Set())).toBe(false);
+  });
+
+  it('one_success fires on first completion', () => {
+    expect(evaluateTriggerRule(['a', 'b'], new Set(['a']), new Set(), new Set(), 'one_success')).toBe(true);
+    expect(evaluateTriggerRule(['a', 'b'], new Set(), new Set(), new Set(), 'one_success')).toBe(false);
+  });
+
+  it('all_done includes skipped deps', () => {
+    expect(evaluateTriggerRule(['a', 'b'], new Set(['a']), new Set(['b']), new Set(), 'all_done')).toBe(true);
+  });
+
+  it('all_success treats excluded deps as satisfied', () => {
+    expect(evaluateTriggerRule(['a', 'b'], new Set(['a']), new Set(), new Set(['b']))).toBe(true);
+  });
+
+  it('defaults to all_success', () => {
+    expect(evaluateTriggerRule(['a'], new Set(['a']), new Set(), new Set())).toBe(true);
+    expect(evaluateTriggerRule(['a'], new Set(), new Set(), new Set())).toBe(false);
+  });
+
+  it('returns true for empty deps', () => {
+    expect(evaluateTriggerRule([], new Set(), new Set(), new Set())).toBe(true);
+  });
+});

apps/coder/src/services/acp-derive.ts

@@ -68,11 +68,18 @@ export function deriveModesFromACP(
 ): { modes: ProviderMode[]; currentModeId: string | null } {
   if (modeState?.availableModes?.length) {
     return {
-      modes: modeState.availableModes.map((mode) => ({
+      // ACP omits the unattended flag; inherit it from the manifest fallback by
-        id: mode.id,
+      // id so the unified permission picker can still detect each agent's bypass
-        label: mode.name,
+      // mode (e.g. opencode `full-access`) from live-probed modes.
-        description: mode.description ?? undefined,
+      modes: modeState.availableModes.map((mode) => {
-      })),
+        const fb = fallbackModes.find((f) => f.id === mode.id);
+        return {
+          id: mode.id,
+          label: mode.name,
+          description: mode.description ?? undefined,
+          ...(fb?.isUnattended ? { isUnattended: true } : {}),
+        };
+      }),
       currentModeId: modeState.currentModeId ?? null,
     };
   }

apps/coder/src/services/arena-analyzer-helpers.ts

@@ -0,0 +1,191 @@
+/**
+ * Pure, side-effect-free helpers for the Arena analyzer.
+ * No DB, no IO, no network — safe to unit-test directly.
+ *
+ * Covers: digest-prompt assembly, judge-prompt assembly, winner extraction
+ * from the judge output, the <2-survivors no-winner rule, and the
+ * cross-examination prompt.
+ */
+
+// ─── Shared types ─────────────────────────────────────────────────────────────
+
+export interface ContestantDigestInput {
+  identity: string;
+  model: string;
+  resultMd: string;
+  diffPatch?: string;
+  benchmarkLine: string;
+}
+
+export interface ContestantDigest {
+  identity: string;
+  model: string;
+  digest: string;
+  benchmarkLine: string;
+}
+
+// ─── Digest stage ─────────────────────────────────────────────────────────────
+
+/**
+ * Build the system + user prompts for the per-contestant digest call.
+ * The digest is a short structured summary; it keeps each call's context small
+ * so the downstream judge only sees digests (not raw diffs).
+ */
+export function buildDigestPrompt(input: ContestantDigestInput): { system: string; user: string } {
+  const system =
+    'You are an expert technical analyst evaluating the output of an AI coding or Q&A battle. ' +
+    'Produce a concise structured digest (under 300 words, Markdown bullet points) covering: ' +
+    '(1) correctness and quality, (2) completeness, (3) notable strengths, (4) notable weaknesses or issues. ' +
+    'Do not reference the battle or other contestants — focus only on this submission.';
+
+  const parts: string[] = [
+    `# Contestant: ${input.identity} / ${input.model}`,
+    `\nBenchmark: ${input.benchmarkLine}`,
+    '\n## Result\n',
+    input.resultMd.slice(0, 8_000),
+  ];
+
+  if (input.diffPatch) {
+    parts.push('\n## Code Changes (diff)\n```diff');
+    parts.push(input.diffPatch.slice(0, 5_000));
+    parts.push('```');
+  }
+
+  return { system, user: parts.join('\n') };
+}
+
+// ─── Judge stage ──────────────────────────────────────────────────────────────
+
+/**
+ * Build the system + user prompts for the comparative judge call.
+ * Receives contestant digests (NOT raw diffs) to keep context bounded.
+ *
+ * The judge output must contain a line starting with WINNER: or NO_WINNER.
+ * The caller extracts it with extractWinner().
+ */
+export function buildJudgePrompt(
+  originalPrompt: string,
+  digests: ContestantDigest[],
+): { system: string; user: string } {
+  const canName = shouldNameWinner(digests.length);
+
+  const winnerInstruction = canName
+    ? 'After your comparative analysis, name the best submission on its own line in this exact format:\n' +
+      'WINNER: <identity>/<model>\n' +
+      'where <identity> and <model> exactly match the heading above. No other text on that line.'
+    : 'Fewer than 2 contestants succeeded. Do NOT name a winner. Write the following on its own line:\nNO_WINNER';
+
+  const system =
+    'You are an expert judge for an AI battle. You have received digest summaries of each ' +
+    "contestant's work on the same task. Write a comparative analysis, then follow these instructions:\n" +
+    winnerInstruction;
+
+  const parts: string[] = [
+    '# Original Task Prompt\n',
+    originalPrompt.slice(0, 2_000),
+    '\n# Contestant Digests\n',
+  ];
+
+  for (const d of digests) {
+    parts.push(`\n## ${d.identity} / ${d.model}`);
+    parts.push(`Benchmark: ${d.benchmarkLine}`);
+    parts.push(d.digest);
+  }
+
+  parts.push(
+    '\n# Instructions\nCompare the contestants and follow the winner-naming instructions above.',
+  );
+
+  return { system, user: parts.join('\n') };
+}
+
+// ─── No-winner rule ───────────────────────────────────────────────────────────
+
+/**
+ * Returns true when enough contestants succeeded to name a winner.
+ * Rule: at least 2 must have produced a result. With 0 or 1 success the
+ * analysis must NOT name a winner (no meaningful comparison possible).
+ */
+export function shouldNameWinner(succeededCount: number): boolean {
+  return succeededCount >= 2;
+}
+
+// ─── Winner extraction ────────────────────────────────────────────────────────
+
+/**
+ * Parse the judge's text output and extract the declared winner.
+ * Looks for a line matching: WINNER: <identity>/<model>
+ * Returns null when no valid winner line is found, or when the line contains
+ * NO_WINNER.
+ *
+ * The parse is lenient on surrounding whitespace and case for the keyword.
+ */
+export function extractWinner(judgeOutput: string): { identity: string; model: string } | null {
+  for (const line of judgeOutput.split('\n')) {
+    const trimmed = line.trim();
+    if (!trimmed.toUpperCase().startsWith('WINNER:')) continue;
+
+    const rest = trimmed.slice('WINNER:'.length).trim();
+    if (rest.toUpperCase() === 'NO_WINNER' || rest === '') return null;
+
+    const slashIdx = rest.indexOf('/');
+    if (slashIdx === -1) return null;
+
+    const identity = rest.slice(0, slashIdx).trim();
+    const model = rest.slice(slashIdx + 1).trim();
+    if (identity && model) return { identity, model };
+  }
+  return null;
+}
+
+// ─── Cross-examination stage ──────────────────────────────────────────────────
+
+/**
+ * Build the system + user prompts for a cross-examination call.
+ * The cross-examiner sees the original prompt, contestant digests, and the
+ * proposed analysis, and is asked to challenge the result.
+ */
+export function buildCrossExamPrompt(opts: {
+  originalPrompt: string;
+  digests: ContestantDigest[];
+  analysisContent: string;
+  proposedWinner: string | null;
+  examinerIdentity: string;
+  examinerModel: string;
+}): { system: string; user: string } {
+  const system =
+    `You are ${opts.examinerIdentity} (model: ${opts.examinerModel}), acting as an independent ` +
+    'cross-examiner in an AI battle. Your role is to critically challenge the proposed analysis ' +
+    'and winner, then give your own verdict. Be rigorous but fair. ' +
+    'End your response with your verdict on its own line:\n' +
+    'VERDICT: <identity>/<model>  — if you agree or disagree with the proposed winner but can name one\n' +
+    'VERDICT: NO_WINNER  — if no clear winner exists';
+
+  const parts: string[] = [
+    '# Original Task Prompt\n',
+    opts.originalPrompt.slice(0, 2_000),
+    '\n# Contestant Digests\n',
+  ];
+
+  for (const d of opts.digests) {
+    parts.push(`\n## ${d.identity} / ${d.model}`);
+    parts.push(`Benchmark: ${d.benchmarkLine}`);
+    parts.push(d.digest);
+  }
+
+  parts.push('\n# Proposed Analysis\n');
+  parts.push(opts.analysisContent.slice(0, 5_000));
+
+  if (opts.proposedWinner) {
+    parts.push(`\n*(Proposed winner: ${opts.proposedWinner})*`);
+  } else {
+    parts.push('\n*(No winner was proposed — fewer than 2 contestants succeeded.)*');
+  }
+
+  parts.push(
+    '\n# Your Cross-Examination\n' +
+      'Challenge the analysis above, then give your independent verdict (VERDICT: … on its own line).',
+  );
+
+  return { system, user: parts.join('\n') };
+}

apps/coder/src/services/arena-analyzer.ts

@@ -0,0 +1,496 @@
+/**
+ * Arena Analyzer — pluggable seam for battle analysis and cross-examination.
+ *
+ * The Analyzer interface is the plug point: a v2 Han Orchestrator flow can
+ * replace the v1 two-stage digest→judge implementation without a schema change.
+ *
+ * v1 implementation uses DEFAULT_MODEL via direct llama-swap calls (arenaModelCall):
+ *   Digest stage  — one call per succeeded contestant, concurrent; produces a
+ *                   bounded summary of each result (result.md + diff.patch for
+ *                   coding, result.md for Q&A).
+ *   Judge stage   — one call with all digests + the original prompt; writes
+ *                   analysis.md, names a winner (unless < 2 succeeded), and
+ *                   updates battles.winner_contestant_id.
+ *
+ * Cross-examination:
+ *   Local model   — direct arenaModelCall to llama-swap with the chosen model.
+ *   Cloud model   — inserts a tasks row (triggers the dispatcher via pg_notify);
+ *                   polls for completion; reads output_summary as the verdict.
+ *   In both cases the verdict is written to cross_examinations.verdict, appended
+ *   to <resultsPath>/cross-exam.md, and a battle_updated frame is published.
+ *
+ * Never throws — all errors are caught, logged, and swallowed so the caller
+ * (arena-runner's onBattleComplete / onCrossExamStart) is never wedged.
+ */
+
+import { readFile, writeFile, mkdir } from 'node:fs/promises';
+import { join } from 'node:path';
+import type { Sql } from '../db.js';
+import type { Broker } from '@boocode/server/broker';
+import type { WsFrame } from '@boocode/contracts/ws-frames';
+import type { FastifyBaseLogger } from 'fastify';
+import type { Config } from '../config.js';
+import type { BattleType } from '@boocode/contracts/arena';
+import { arenaModelCall } from './arena-model-call.js';
+import {
+  buildDigestPrompt,
+  buildJudgePrompt,
+  buildCrossExamPrompt,
+  extractWinner,
+  shouldNameWinner,
+  type ContestantDigest,
+} from './arena-analyzer-helpers.js';
+
+// ─── Public interface ─────────────────────────────────────────────────────────
+
+/** Pluggable analysis seam — swap to a Han Orchestrator flow in v2. */
+export interface Analyzer {
+  /** Run the two-stage digest→judge analysis for a completed battle. */
+  analyze(battleId: string): Promise<void>;
+  /**
+   * Run a cross-examination for an already-inserted cross_examinations row.
+   * The result is written back to that row and a battle_updated frame is published.
+   */
+  crossExamine(
+    battleId: string,
+    crossExamId: string,
+    opts: { identity: string; model: string },
+  ): Promise<void>;
+}
+
+// ─── Internal DB row types ────────────────────────────────────────────────────
+
+interface BattleRow {
+  id: string;
+  project_id: string;
+  battle_type: BattleType;
+  prompt: string;
+  status: string;
+  results_path: string | null;
+  winner_contestant_id: string | null;
+}
+
+interface ContestantRow {
+  id: string;
+  identity: string;
+  model: string;
+  lane: string;
+  status: string;
+  result_path: string | null;
+  duration_ms: number | null;
+  tokens_per_sec: number | null;
+}
+
+// ─── Factory ──────────────────────────────────────────────────────────────────
+
+interface AnalyzerDeps {
+  sql: Sql;
+  broker: Broker;
+  log: FastifyBaseLogger;
+  config: Pick<Config, 'LLAMA_SWAP_URL' | 'DEFAULT_MODEL'>;
+  /** Model IDs served by local llama-swap — cross-exam routing uses this. */
+  localModels: ReadonlySet<string>;
+}
+
+export function createAnalyzer(deps: AnalyzerDeps): Analyzer {
+  const { sql, broker, log, config, localModels } = deps;
+
+  // ─── analyze ──────────────────────────────────────────────────────────────
+
+  async function analyze(battleId: string): Promise<void> {
+    try {
+      await runAnalysis(battleId);
+    } catch (err) {
+      log.error(
+        { err: errMsg(err), battleId },
+        'arena-analyzer: analysis failed',
+      );
+    }
+  }
+
+  async function runAnalysis(battleId: string): Promise<void> {
+    const battle = await loadBattle(battleId);
+    if (!battle) {
+      log.warn({ battleId }, 'arena-analyzer: battle not found');
+      return;
+    }
+
+    const contestants = await loadContestants(battleId);
+    const succeeded = contestants.filter((c) => c.status === 'done' && c.result_path);
+
+    log.info(
+      { battleId, total: contestants.length, succeeded: succeeded.length },
+      'arena-analyzer: starting analysis',
+    );
+
+    // Digest stage — concurrent, one call per succeeded contestant.
+    const digests = (
+      await Promise.all(succeeded.map((c) => digestContestant(battle, c)))
+    ).filter((d): d is ContestantDigest => d !== null);
+
+    // Failed contestants are noted in the analysis even if they produced no digest.
+    const failedNotes = contestants
+      .filter((c) => c.status === 'error')
+      .map((c) => `- **${c.identity} / ${c.model}**: failed (no result)\n`);
+
+    // Judge stage — single call with all digests.
+    const { analysisText, winner } = await judgeContestants(battle, digests, failedNotes);
+
+    // Write analysis.md to the battle results folder.
+    const resultsPath = battle.results_path;
+    if (resultsPath) {
+      await mkdir(resultsPath, { recursive: true });
+      await writeFile(join(resultsPath, 'analysis.md'), analysisText, 'utf8');
+    }
+
+    // Resolve the winner to a contestant id and update the battle row.
+    let winnerId: string | null = null;
+    if (winner && shouldNameWinner(succeeded.length)) {
+      const winnerContestant = contestants.find(
+        (c) => c.identity === winner.identity && c.model === winner.model,
+      );
+      if (winnerContestant) {
+        winnerId = winnerContestant.id;
+        await sql`
+          UPDATE battles
+          SET winner_contestant_id = ${winnerId}, updated_at = clock_timestamp()
+          WHERE id = ${battleId}
+        `;
+        log.info({ battleId, winnerId, identity: winner.identity, model: winner.model }, 'arena-analyzer: winner set');
+      } else {
+        log.warn({ battleId, winner }, 'arena-analyzer: judge named a winner not found in contestants');
+      }
+    }
+
+    publishUser({
+      type: 'battle_updated',
+      battle_id: battleId,
+      winner_contestant_id: winnerId,
+      analysis_ready: true,
+    });
+
+    log.info({ battleId }, 'arena-analyzer: analysis complete');
+  }
+
+  // ─── crossExamine ─────────────────────────────────────────────────────────
+
+  async function crossExamine(
+    battleId: string,
+    crossExamId: string,
+    opts: { identity: string; model: string },
+  ): Promise<void> {
+    try {
+      await runCrossExam(battleId, crossExamId, opts);
+    } catch (err) {
+      log.error(
+        { err: errMsg(err), battleId, crossExamId },
+        'arena-analyzer: cross-exam failed',
+      );
+    }
+  }
+
+  async function runCrossExam(
+    battleId: string,
+    crossExamId: string,
+    opts: { identity: string; model: string },
+  ): Promise<void> {
+    const battle = await loadBattle(battleId);
+    if (!battle) {
+      log.warn({ battleId }, 'arena-analyzer: battle not found for cross-exam');
+      return;
+    }
+
+    const contestants = await loadContestants(battleId);
+
+    // Re-read the digests (if contestants have results) for context.
+    const succeeded = contestants.filter((c) => c.status === 'done' && c.result_path);
+    const digests = (
+      await Promise.all(succeeded.map((c) => digestContestant(battle, c)))
+    ).filter((d): d is ContestantDigest => d !== null);
+
+    // Read analysis.md for the proposed analysis content.
+    let analysisContent = '';
+    if (battle.results_path) {
+      analysisContent = await readFile(
+        join(battle.results_path, 'analysis.md'), 'utf8',
+      ).catch(() => '');
+    }
+
+    // Resolve proposed winner label.
+    let proposedWinner: string | null = null;
+    if (battle.winner_contestant_id) {
+      const w = contestants.find((c) => c.id === battle.winner_contestant_id);
+      if (w) proposedWinner = `${w.identity}/${w.model}`;
+    }
+
+    const { system, user } = buildCrossExamPrompt({
+      originalPrompt: battle.prompt,
+      digests,
+      analysisContent,
+      proposedWinner,
+      examinerIdentity: opts.identity,
+      examinerModel: opts.model,
+    });
+
+    log.info({ battleId, crossExamId, identity: opts.identity, model: opts.model }, 'arena-analyzer: running cross-exam');
+
+    const verdict = await executeModelCall({
+      battleId,
+      projectId: battle.project_id,
+      identity: opts.identity,
+      model: opts.model,
+      system,
+      user,
+    });
+
+    // Persist verdict and append to cross-exam.md.
+    await sql`
+      UPDATE cross_examinations
+      SET verdict = ${verdict}
+      WHERE id = ${crossExamId}
+    `;
+
+    if (battle.results_path) {
+      const crossExamPath = join(battle.results_path, 'cross-exam.md');
+      const section =
+        `\n---\n\n# Cross-Examination by ${opts.identity} / ${opts.model}\n\n` +
+        `${verdict}\n`;
+      await writeFile(crossExamPath, section, { flag: 'a', encoding: 'utf8' });
+    }
+
+    publishUser({
+      type: 'battle_updated',
+      battle_id: battleId,
+      cross_exam_id: crossExamId,
+    });
+
+    log.info({ battleId, crossExamId }, 'arena-analyzer: cross-exam complete');
+  }
+
+  // ─── Model call routing ───────────────────────────────────────────────────
+
+  /**
+   * Route a one-shot model call to llama-swap (local) or the task dispatcher
+   * (cloud). Cloud dispatch inserts a tasks row and polls for completion.
+   */
+  async function executeModelCall(opts: {
+    battleId: string;
+    projectId: string;
+    identity: string;
+    model: string;
+    system: string;
+    user: string;
+  }): Promise<string> {
+    const isLocal = localModels.has(opts.model) || localModels.has(`llama-swap/${opts.model}`);
+
+    if (isLocal) {
+      return arenaModelCall({
+        config,
+        model: opts.model,
+        system: opts.system,
+        user: opts.user,
+        maxTokens: 2_000,
+        temperature: 0.3,
+      });
+    }
+
+    // Cloud path: dispatch through the task system and poll for completion.
+    return executeCloudModelCall(opts);
+  }
+
+  async function executeCloudModelCall(opts: {
+    projectId: string;
+    identity: string;
+    model: string;
+    system: string;
+    user: string;
+  }): Promise<string> {
+    // The cross-exam prompt is the full input to the external agent. We embed
+    // the system prompt as a preamble in the user message (external agents don't
+    // take a separate system arg through the tasks dispatcher).
+    const input = `${opts.system}\n\n${opts.user}`;
+
+    // For well-known external agents, stamp the agent name so the dispatcher
+    // routes via PTY/ACP. For unknown identities fall back to native inference
+    // (agent = null → DEFAULT_MODEL text generation).
+    const knownAgents = new Set(['claude', 'opencode', 'qwen', 'goose']);
+    const agentName = knownAgents.has(opts.identity) ? opts.identity : null;
+
+    const [task] = await sql<{ id: string }[]>`
+      INSERT INTO tasks (project_id, input, agent, model)
+      VALUES (${opts.projectId}, ${input}, ${agentName}, ${opts.model})
+      RETURNING id
+    `;
+    const taskId = task!.id;
+
+    log.info({ taskId, identity: opts.identity, model: opts.model }, 'arena-analyzer: cloud cross-exam task dispatched');
+
+    // Poll until terminal (up to 5 minutes).
+    const timeoutMs = 5 * 60 * 1_000;
+    const pollMs = 2_000;
+    const deadline = Date.now() + timeoutMs;
+
+    while (Date.now() < deadline) {
+      await sleep(pollMs);
+      const [row] = await sql<{ state: string; output_summary: string | null }[]>`
+        SELECT state, output_summary FROM tasks WHERE id = ${taskId}
+      `;
+      if (!row) break;
+      if (row.state === 'completed') return row.output_summary ?? '';
+      if (row.state === 'failed' || row.state === 'cancelled') {
+        throw new Error(`cross-exam task ${row.state}: ${row.output_summary ?? ''}`);
+      }
+    }
+
+    throw new Error(`cloud cross-exam task timed out after ${timeoutMs / 1000}s`);
+  }
+
+  // ─── Digest helper ────────────────────────────────────────────────────────
+
+  async function digestContestant(
+    battle: BattleRow,
+    c: ContestantRow,
+  ): Promise<ContestantDigest | null> {
+    if (!c.result_path) return null;
+
+    const resultMd = await readFile(join(c.result_path, 'result.md'), 'utf8').catch(() => '');
+
+    let diffPatch: string | undefined;
+    if (battle.battle_type === 'coding') {
+      diffPatch = await readFile(join(c.result_path, 'diff.patch'), 'utf8').catch(
+        () => undefined,
+      );
+    }
+
+    const benchmarkLine = formatBenchmarkLine(c);
+    const { system, user } = buildDigestPrompt({
+      identity: c.identity,
+      model: c.model,
+      resultMd,
+      diffPatch,
+      benchmarkLine,
+    });
+
+    let digest: string;
+    try {
+      digest = await arenaModelCall({
+        config,
+        model: config.DEFAULT_MODEL,
+        system,
+        user,
+        maxTokens: 500,
+        temperature: 0.3,
+      });
+    } catch (err) {
+      log.warn(
+        { err: errMsg(err), identity: c.identity, model: c.model },
+        'arena-analyzer: digest call failed — skipping contestant',
+      );
+      return null;
+    }
+
+    return { identity: c.identity, model: c.model, digest, benchmarkLine };
+  }
+
+  // ─── Judge helper ─────────────────────────────────────────────────────────
+
+  async function judgeContestants(
+    battle: BattleRow,
+    digests: ContestantDigest[],
+    failedNotes: string[],
+  ): Promise<{ analysisText: string; winner: { identity: string; model: string } | null }> {
+    const { system, user } = buildJudgePrompt(battle.prompt, digests);
+
+    let judgeOutput = '';
+    try {
+      judgeOutput = await arenaModelCall({
+        config,
+        model: config.DEFAULT_MODEL,
+        system,
+        user,
+        maxTokens: 2_000,
+        temperature: 0.3,
+      });
+    } catch (err) {
+      log.error({ err: errMsg(err), battleId: battle.id }, 'arena-analyzer: judge call failed');
+      judgeOutput = '*(Judge call failed — no comparison produced.)*';
+    }
+
+    const winner = shouldNameWinner(digests.length) ? extractWinner(judgeOutput) : null;
+
+    const sections: string[] = [
+      `# Arena Analysis`,
+      `\n**Battle type:** ${battle.battle_type}`,
+    ];
+
+    if (failedNotes.length > 0) {
+      sections.push('\n## Failed Contestants\n');
+      sections.push(...failedNotes);
+    }
+
+    if (digests.length > 0) {
+      sections.push('\n## Contestant Digests\n');
+      for (const d of digests) {
+        sections.push(`### ${d.identity} / ${d.model}`);
+        sections.push(`*Benchmark: ${d.benchmarkLine}*\n`);
+        sections.push(d.digest);
+      }
+    }
+
+    sections.push("\n## Judge's Verdict\n");
+    sections.push(judgeOutput);
+
+    if (winner) {
+      sections.push(`\n## Winner\n**${winner.identity} / ${winner.model}**`);
+    } else {
+      const reason =
+        digests.length < 2
+          ? 'fewer than 2 contestants produced results'
+          : 'no clear winner identified';
+      sections.push(`\n## Winner\n*No winner named (${reason}).*`);
+    }
+
+    return { analysisText: sections.join('\n'), winner };
+  }
+
+  // ─── DB helpers ───────────────────────────────────────────────────────────
+
+  async function loadBattle(battleId: string): Promise<BattleRow | null> {
+    const [b] = await sql<BattleRow[]>`
+      SELECT id, project_id, battle_type, prompt, status, results_path, winner_contestant_id
+      FROM battles WHERE id = ${battleId}
+    `;
+    return b ?? null;
+  }
+
+  async function loadContestants(battleId: string): Promise<ContestantRow[]> {
+    return sql<ContestantRow[]>`
+      SELECT id, identity, model, lane, status, result_path, duration_ms, tokens_per_sec
+      FROM contestants WHERE battle_id = ${battleId}
+      ORDER BY created_at ASC
+    `;
+  }
+
+  // ─── Misc helpers ─────────────────────────────────────────────────────────
+
+  function formatBenchmarkLine(c: ContestantRow): string {
+    const parts: string[] = [];
+    if (c.duration_ms !== null) parts.push(`${c.duration_ms}ms`);
+    if (c.tokens_per_sec !== null) parts.push(`${c.tokens_per_sec.toFixed(1)} tok/s`);
+    return parts.length > 0 ? parts.join(', ') : 'no benchmark';
+  }
+
+  function publishUser(frame: Record<string, unknown>): void {
+    broker.publishUserFrame('default', frame as unknown as WsFrame);
+  }
+
+  function sleep(ms: number): Promise<void> {
+    return new Promise((resolve) => setTimeout(resolve, ms));
+  }
+
+  return { analyze, crossExamine };
+}
+
+function errMsg(e: unknown): string {
+  return e instanceof Error ? e.message : String(e);
+}

apps/coder/src/services/arena-decisions.ts

@@ -0,0 +1,188 @@
+/**
+ * Pure scheduling and classification decisions for the Arena battle-runner.
+ * No database, no IO. Mirrors the pattern of flow-runner-decisions.ts.
+ *
+ * Vocabulary:
+ *   local lane  — llama-swap-backed contestants, run strictly one at a time
+ *   cloud lane  — cloud-backed contestants, run all in parallel
+ *
+ * A contestant's status lifecycle:
+ *   queued → running → done | error
+ */
+import type { BattleType, ContestantLane, TokenBreakdown } from '@boocode/contracts/arena';
+
+// ─── Lane classification ──────────────────────────────────────────────────────
+
+/**
+ * Classify a contestant into a lane.
+ *
+ * Q&A contestants always run on the native (llama-swap) backend → local.
+ * Coding contestants: their MODEL is checked against the localModels set
+ * (all model IDs served by the local llama-swap server). This means an
+ * opencode or qwen contestant pointed at a local model counts as local,
+ * which correctly captures GPU-contention and fair benchmarking (ADR 0001).
+ *
+ * @param battleType  'coding' | 'qa'
+ * @param identity    backend name (coding) or persona name (qa) — not used for lane logic
+ * @param model       the contestant's model id
+ * @param localModels set of model IDs served by the local llama-swap server
+ */
+export function classifyLane(
+  battleType: BattleType,
+  _identity: string,
+  model: string,
+  localModels: ReadonlySet<string>,
+): ContestantLane {
+  if (battleType === 'qa') return 'local';
+  return localModels.has(model) ? 'local' : 'cloud';
+}
+
+// ─── Local-lane queue ─────────────────────────────────────────────────────────
+
+export interface ContestantSlot {
+  id: string;
+  lane: ContestantLane;
+  status: string;
+}
+
+/**
+ * The next queued local contestant to dispatch — the first 'queued' contestant
+ * in the local lane, in creation order (caller must supply rows in created_at ASC).
+ * Returns null when the local queue is empty or all local slots are non-queued.
+ */
+export function nextLocalContestant(contestants: readonly ContestantSlot[]): string | null {
+  for (const c of contestants) {
+    if (c.lane === 'local' && c.status === 'queued') return c.id;
+  }
+  return null;
+}
+
+// ─── Battle completion ────────────────────────────────────────────────────────
+
+/**
+ * True when every contestant has reached a terminal state (done | error).
+ * Returns false for an empty list — a battle with no contestants never completes.
+ */
+export function isBattleComplete(contestants: readonly { status: string }[]): boolean {
+  if (contestants.length === 0) return false;
+  return contestants.every((c) => c.status === 'done' || c.status === 'error');
+}
+
+// ─── Benchmark ────────────────────────────────────────────────────────────────
+
+export interface Benchmark {
+  durationMs: number;
+  tokensPerSec: number | null;
+  tokenBreakdown: TokenBreakdown | null;
+}
+
+/**
+ * Compute the benchmark for a contestant.
+ * Wall-clock duration is captured for every contestant; tokens/sec is only
+ * meaningful for local (llama-swap) contestants where the model has sole
+ * access to the GPU and the measurement is fair.
+ */
+export function computeBenchmark(
+  startedAt: Date,
+  endedAt: Date,
+  costTokens: number | null,
+  lane: ContestantLane,
+  tokenBreakdown: TokenBreakdown | null = null,
+): Benchmark {
+  const durationMs = Math.max(0, endedAt.getTime() - startedAt.getTime());
+  const tokensPerSec =
+    lane === 'local' && costTokens !== null && durationMs > 0
+      ? (costTokens / durationMs) * 1000
+      : null;
+  return { durationMs, tokensPerSec, tokenBreakdown };
+}
+
+// ─── Slug / path helpers ──────────────────────────────────────────────────────
+
+/**
+ * Sanitize a string for use as a directory name component.
+ * Lowercases, replaces non-alphanumeric runs with '-', trims leading/trailing
+ * dashes, and caps at 64 characters.
+ */
+export function sanitizeSlug(s: string): string {
+  return s
+    .toLowerCase()
+    .replace(/[^a-z0-9]+/g, '-')
+    .replace(/^-+|-+$/g, '')
+    .slice(0, 64);
+}
+
+/**
+ * Build the dated battle slug used as the Arena results folder name.
+ * Format: YYYY-MM-DD-<battleType>-<first-8-hex-of-uuid>
+ * Deterministic: callers can rebuild it from (id, type, created_at) on resume.
+ */
+export function buildBattleSlug(battleId: string, battleType: BattleType, createdAt: Date): string {
+  const date = createdAt.toISOString().slice(0, 10);
+  const shortId = battleId.replace(/-/g, '').slice(0, 8);
+  return `${date}-${battleType}-${shortId}`;
+}
+
+/**
+ * Build the per-contestant results directory name within a battle folder.
+ * Format: <sanitized-identity>-<sanitized-model>
+ */
+export function buildContestantDir(identity: string, model: string): string {
+  return `${sanitizeSlug(identity)}-${sanitizeSlug(model)}`;
+}
+
+// ─── Resume reconciliation ────────────────────────────────────────────────────
+
+export type ContestantResumeAction =
+  | 'keep'
+  | 're-dispatch'
+  | 'mark-done'
+  | 'mark-error'
+  | 'mark-cancelled';
+
+export interface ContestantResumeDecision {
+  contestantId: string;
+  action: ContestantResumeAction;
+}
+
+/**
+ * Decide what to do with ONE contestant during startup resume.
+ * Mirrors reconcileResumeStep from flow-runner-decisions.ts.
+ *
+ * @param status    contestants.status
+ * @param taskId    contestants.task_id (null when not yet dispatched)
+ * @param taskState tasks.state for taskId, or null if the task row is absent
+ */
+export function reconcileContestantResume(
+  status: string,
+  taskId: string | null,
+  taskState: string | null,
+): ContestantResumeAction {
+  if (status !== 'running') return 'keep';
+  if (!taskId || taskState === null) return 're-dispatch';
+  switch (taskState) {
+    case 'completed': return 'mark-done';
+    case 'failed':    return 'mark-error';
+    case 'cancelled': return 'mark-cancelled';
+    case 'pending':   return 'keep'; // dispatcher startup poll will run it normally
+    default:          return 're-dispatch'; // 'running'/'blocked' — process is dead
+  }
+}
+
+/**
+ * Reconcile every contestant of an in-flight battle for startup resume.
+ * Returns one decision per contestant. Pure — no IO.
+ */
+export function reconcileContestants(
+  contestants: ReadonlyArray<{ contestantId: string; taskId: string | null; status: string }>,
+  taskStates: ReadonlyMap<string, string>,
+): ContestantResumeDecision[] {
+  return contestants.map((c) => ({
+    contestantId: c.contestantId,
+    action: reconcileContestantResume(
+      c.status,
+      c.taskId,
+      c.taskId ? (taskStates.get(c.taskId) ?? null) : null,
+    ),
+  }));
+}

apps/coder/src/services/arena-model-call.ts

@@ -0,0 +1,70 @@
+/**
+ * One-shot model completion for the Arena analyzer.
+ *
+ * Calls the local llama-swap server directly for a single non-streaming
+ * completion. Used for the digest and judge stages (always DEFAULT_MODEL)
+ * and for local-model cross-examinations (any local model).
+ *
+ * Mirrors apps/server/src/services/task-model.ts but targets the coder's
+ * config shape and uses a longer timeout appropriate for analysis calls.
+ */
+
+import type { Config } from '../config.js';
+
+const TIMEOUT_MS = 120_000;
+
+export async function arenaModelCall(opts: {
+  config: Pick<Config, 'LLAMA_SWAP_URL'>;
+  model: string;
+  system: string;
+  user: string;
+  maxTokens?: number;
+  temperature?: number;
+}): Promise<string> {
+  const { config, model, system, user } = opts;
+  const maxTokens = opts.maxTokens ?? 2_000;
+  const temperature = opts.temperature ?? 0.3;
+
+  const res = await fetch(`${config.LLAMA_SWAP_URL}/v1/chat/completions`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      model,
+      messages: [
+        { role: 'system', content: system },
+        { role: 'user', content: user },
+      ],
+      max_tokens: maxTokens,
+      temperature,
+      stream: false,
+      chat_template_kwargs: { enable_thinking: false },
+    }),
+    signal: AbortSignal.timeout(TIMEOUT_MS),
+  });
+
+  if (!res.ok) {
+    const text = await res.text().catch(() => '');
+    throw new Error(`llama-swap responded ${res.status}: ${text.slice(0, 200)}`);
+  }
+
+  const data = (await res.json()) as {
+    choices?: Array<{
+      message?: { content?: string; reasoning_content?: string };
+    }>;
+  };
+
+  const choice = data.choices?.[0]?.message;
+  if (!choice) return '';
+
+  const content = (choice.content ?? '').trim();
+  if (content.length > 0) return content;
+
+  // For thinking-mode models the answer sometimes only lands in reasoning_content.
+  const reasoning = (choice.reasoning_content ?? '').trim();
+  if (reasoning.length > 0) {
+    const lines = reasoning.split('\n').filter((l) => l.trim().length > 0);
+    return lines[lines.length - 1] ?? '';
+  }
+
+  return '';
+}

apps/coder/src/services/arena-runner.ts

@@ -0,0 +1,895 @@
+/**
+ * Arena battle-runner — DB-backed execution engine for Arena battles.
+ *
+ * Mirrors flow-runner.ts but implements the Arena's two-lane scheduler instead
+ * of the Orchestrator's wave scheduler. Persists to battles/contestants tables
+ * (not flow_runs/flow_steps). Each contestant is dispatched as a real tasks row
+ * via an injected DispatchContestantFn (Phase 4 wires this to the dispatcher).
+ * Advances on the dispatcher's onTaskTerminal hook.
+ *
+ * Scheduling:
+ *   - Cloud lane: all contestants start immediately, in parallel.
+ *   - Local lane: contestants run strictly one at a time (serial queue). Only
+ *     the first local contestant runs at start; the next is dispatched when the
+ *     current one terminates. Both lanes run concurrently with each other.
+ *
+ * Results:
+ *   Written to <projectRoot>/Arena/<battleSlug>/<identity>-<model>/
+ *   Coding: result.md + diff.patch (from the contestant's worktree).
+ *   Q&A:    result.md with the text answer.
+ *
+ * Analyzer seam:
+ *   onBattleComplete is called when all contestants are terminal. Phase 5 wires
+ *   this to the two-stage digest→judge analyzer. A failed contestant does NOT
+ *   abort the battle — others continue and the analyzer judges survivors.
+ */
+import type { Sql } from '../db.js';
+import type { Broker } from '@boocode/server/broker';
+import type { WsFrame } from '@boocode/contracts/ws-frames';
+import type { FastifyBaseLogger } from 'fastify';
+import type { BattleType, ContestantLane } from '@boocode/contracts/arena';
+import { mkdir, writeFile } from 'node:fs/promises';
+import { join } from 'node:path';
+import { diffWorktree } from './worktrees.js';
+import {
+  buildBattleSlug,
+  buildContestantDir,
+  classifyLane,
+  computeBenchmark,
+  isBattleComplete,
+  nextLocalContestant,
+  reconcileContestants,
+  type ContestantResumeAction,
+  type ContestantSlot,
+} from './arena-decisions.js';
+
+// ─── Public types ─────────────────────────────────────────────────────────────
+
+export interface ContestantSpec {
+  /** Backend name (coding) or persona name (qa). */
+  identity: string;
+  model: string;
+}
+
+export interface BattleStartOpts {
+  projectId: string;
+  battleType: BattleType;
+  prompt: string;
+  /** 2–6 contestants. Duplicate (identity, model) pairs are rejected by the schema UNIQUE constraint. */
+  contestants: ContestantSpec[];
+}
+
+/**
+ * Injected dispatch function — Phase 4 wires this to the real task inserter.
+ * Must INSERT a tasks row and return its id. The arena-runner sets the
+ * contestant's task_id and status after this call.
+ * `sessionId` is returned when already known (Q&A pre-creates the session);
+ * null for coding contestants whose session is created lazily by the dispatcher.
+ */
+export type DispatchContestantFn = (opts: {
+  projectId: string;
+  contestantId: string;
+  prompt: string;
+  identity: string;
+  model: string;
+  battleType: BattleType;
+}) => Promise<{ taskId: string; sessionId: string | null }>;
+
+/**
+ * Called once when every contestant in a battle has reached a terminal state.
+ * Phase 5 wires this to the two-stage digest→judge analyzer.
+ * Must never throw — the caller swallows errors.
+ */
+export type OnBattleComplete = (battleId: string) => void;
+
+/**
+ * Called after a cross_examinations row has been inserted, with its id.
+ * Phase 5 wires this to the analyzer's cross-examination runner.
+ * Must never throw — the caller swallows errors.
+ */
+export type OnCrossExamStart = (opts: {
+  battleId: string;
+  crossExamId: string;
+  identity: string;
+  model: string;
+}) => void;
+
+export interface BattleRunner {
+  /** Start a battle: persist it + its contestants, classify lanes, dispatch initial wave. */
+  startBattle(opts: BattleStartOpts): Promise<{ battleId: string }>;
+  /**
+   * Wire to createDispatcher({ onTaskTerminal }). Fires when ANY task settles;
+   * the runner ignores tasks it doesn't own. Never throws.
+   */
+  handleTaskTerminal(taskId: string, state: string): void;
+  /**
+   * Re-advance any battles still marked 'running' after a coder restart.
+   * Mirrors flow-runner's initResume (D-9). Never throws.
+   */
+  initResume(): Promise<void>;
+  /**
+   * Cancel a running battle. Marks it and all non-terminal contestants cancelled,
+   * publishes frames, and returns the task_ids of in-flight contestants so the
+   * route can abort them via the dispatcher's cancelExternalTask.
+   */
+  cancelBattle(battleId: string): Promise<{ cancelled: boolean; taskIds: string[] }>;
+  /**
+   * Trigger analysis for a completed (or manually re-analyzed) battle.
+   * Phase 5 wires this to the two-stage digest→judge analyzer. For now, calls
+   * the injected onBattleComplete seam directly.
+   */
+  triggerAnalysis(battleId: string): Promise<{ triggered: boolean }>;
+  /**
+   * Start a cross-examination on a battle. Inserts a cross_examinations row and
+   * invokes the analyzer seam. Phase 5 fills the actual verdict logic.
+   */
+  startCrossExam(
+    battleId: string,
+    opts: { identity: string; model: string },
+  ): Promise<{ crossExamId: string }>;
+  /**
+   * Manually set (or clear) the winner. Validates the contestant belongs to the
+   * battle, updates battles.winner_contestant_id, and publishes a battle_updated
+   * frame so the pane reflects the override immediately.
+   */
+  setWinner(battleId: string, winnerId: string | null): Promise<{
+    ok: boolean;
+    notFound?: boolean;
+    invalidContestant?: boolean;
+  }>;
+}
+
+// ─── Internal row shapes ──────────────────────────────────────────────────────
+
+interface ContestantRow {
+  id: string;
+  battle_id: string;
+  identity: string;
+  model: string;
+  lane: ContestantLane;
+  task_id: string | null;
+  worktree_id: string | null;
+  status: string;
+}
+
+interface BattleRow {
+  id: string;
+  project_id: string;
+  battle_type: BattleType;
+  prompt: string;
+  status: string;
+  results_path: string | null;
+  created_at: Date;
+}
+
+// ─── Deps / factory ───────────────────────────────────────────────────────────
+
+interface Deps {
+  sql: Sql;
+  broker: Broker;
+  log: FastifyBaseLogger;
+  dispatch: DispatchContestantFn;
+  onBattleComplete: OnBattleComplete;
+  /**
+   * Called after a cross_examinations row is inserted. Phase 5 wires this to
+   * the analyzer's cross-examination runner. Optional: absent → no cross-exam
+   * logic runs (stub behaviour for tests).
+   */
+  onCrossExamStart?: OnCrossExamStart;
+  /**
+   * Model IDs served by the local llama-swap server. Used for lane classification:
+   * a contestant whose model is in this set runs in the local lane (serial, GPU-fair).
+   * Q&A contestants are always local regardless of this set.
+   * Defaults to an empty set → all coding contestants go to the cloud lane.
+   */
+  localModels?: ReadonlySet<string>;
+}
+
+const DEFAULT_LOCAL_MODELS: ReadonlySet<string> = new Set();
+
+export function createBattleRunner(deps: Deps): BattleRunner {
+  const { sql, broker, log, dispatch, onBattleComplete, onCrossExamStart } = deps;
+  const localModels = deps.localModels ?? DEFAULT_LOCAL_MODELS;
+
+  // Serialize local-lane advance per battle so two near-simultaneous terminal
+  // callbacks don't double-dispatch the next local contestant.
+  const advanceChain = new Map<string, Promise<void>>();
+
+  // Delta bridge: per-contestant broker unsubscribe functions.
+  // 'terminated' sentinel prevents a late-arriving setupDeltaBridge from
+  // registering a subscription that would never be cleaned up.
+  const deltaUnsubs = new Map<string, (() => void) | 'terminated'>();
+
+  function publishUser(frame: Record<string, unknown>): void {
+    broker.publishUserFrame('default', frame as unknown as WsFrame);
+  }
+
+  /**
+   * Subscribe to the contestant's inference session and forward delta frames
+   * to the user channel as contestant_updated{delta}. Polls for session_id
+   * when not immediately known (coding contestants whose session is created
+   * lazily by the dispatcher). Unsubscribes on termination or max retries.
+   */
+  async function setupDeltaBridge(
+    battleId: string,
+    contestantId: string,
+    taskId: string,
+    knownSessionId: string | null,
+  ): Promise<void> {
+    let sessionId = knownSessionId;
+    if (!sessionId) {
+      // Coding contestant: session_id is written by the dispatcher just before
+      // inference starts. Poll until it appears or the contestant terminates.
+      for (let i = 0; i < 50; i++) {
+        if (deltaUnsubs.get(contestantId) === 'terminated') return;
+        const [row] = await sql<{ session_id: string | null }[]>`
+          SELECT session_id FROM tasks WHERE id = ${taskId}
+        `.catch(() => []);
+        if (row?.session_id) { sessionId = row.session_id; break; }
+        await new Promise((r) => setTimeout(r, 200));
+      }
+    }
+    if (!sessionId) return;
+    if (deltaUnsubs.get(contestantId) === 'terminated') return;
+
+    const unsub = broker.subscribe(sessionId, (frame) => {
+      if (frame.type === 'delta') {
+        const deltaContent = (frame as unknown as { content?: unknown }).content;
+        if (typeof deltaContent === 'string') {
+          publishUser({
+            type: 'contestant_updated',
+            battle_id: battleId,
+            contestant_id: contestantId,
+            delta: deltaContent,
+          });
+        }
+      }
+    });
+
+    const existing = deltaUnsubs.get(contestantId);
+    if (existing === 'terminated') {
+      unsub();
+    } else {
+      deltaUnsubs.set(contestantId, unsub);
+    }
+  }
+
+  function teardownDeltaBridge(contestantId: string): void {
+    const entry = deltaUnsubs.get(contestantId);
+    if (typeof entry === 'function') {
+      entry();
+      deltaUnsubs.delete(contestantId);
+    } else {
+      deltaUnsubs.set(contestantId, 'terminated');
+    }
+  }
+
+  // ─── startBattle ────────────────────────────────────────────────────────────
+
+  async function startBattle(opts: BattleStartOpts): Promise<{ battleId: string }> {
+    if (opts.contestants.length < 2 || opts.contestants.length > 6) {
+      throw new Error(`battle requires 2–6 contestants; got ${opts.contestants.length}`);
+    }
+
+    const [proj] = await sql<{ path: string }[]>`SELECT path FROM projects WHERE id = ${opts.projectId}`;
+    if (!proj) throw new Error(`project not found: ${opts.projectId}`);
+
+    // Insert the battle row as 'running'; update results_path once we have the id.
+    const [battle] = await sql<{ id: string; created_at: Date }[]>`
+      INSERT INTO battles (project_id, battle_type, prompt, status)
+      VALUES (${opts.projectId}, ${opts.battleType}, ${opts.prompt}, 'running')
+      RETURNING id, created_at
+    `;
+    const battleId = battle!.id;
+    const battleSlug = buildBattleSlug(battleId, opts.battleType, battle!.created_at);
+    const resultsPath = join(proj.path, 'Arena', battleSlug);
+
+    await sql`
+      UPDATE battles SET results_path = ${resultsPath}, updated_at = clock_timestamp()
+      WHERE id = ${battleId}
+    `;
+
+    // Insert all contestant rows with lane classification.
+    const contestantRows: Array<{ id: string; identity: string; model: string; lane: ContestantLane }> = [];
+    for (const spec of opts.contestants) {
+      const lane = classifyLane(opts.battleType, spec.identity, spec.model, localModels);
+      const [row] = await sql<{ id: string }[]>`
+        INSERT INTO contestants (battle_id, identity, model, lane, status)
+        VALUES (${battleId}, ${spec.identity}, ${spec.model}, ${lane}, 'queued')
+        RETURNING id
+      `;
+      contestantRows.push({ id: row!.id, identity: spec.identity, model: spec.model, lane });
+    }
+
+    // Write initial manifest so the results folder is always populated.
+    await writeManifest(
+      battleId, resultsPath, opts.battleType, opts.prompt, battle!.created_at,
+      contestantRows.map((c) => ({ identity: c.identity, model: c.model, lane: c.lane })),
+      null,
+    ).catch((err) => {
+      log.warn({ err: errMsg(err), battleId }, 'arena-runner: initial manifest write failed');
+    });
+
+    publishUser({
+      type: 'battle_started',
+      battle_id: battleId,
+      battle_type: opts.battleType,
+      prompt: opts.prompt,
+      contestants: contestantRows.map((c) => ({
+        id: c.id,
+        identity: c.identity,
+        model: c.model,
+        lane: c.lane,
+      })),
+    });
+
+    // Dispatch: cloud lane starts all contestants in parallel; local lane starts
+    // only the first queued contestant (serial queue).
+    let localStarted = false;
+    for (const c of contestantRows) {
+      if (c.lane === 'cloud') {
+        await dispatchContestant(battleId, opts.projectId, opts.battleType, opts.prompt, c);
+      } else if (!localStarted) {
+        await dispatchContestant(battleId, opts.projectId, opts.battleType, opts.prompt, c);
+        localStarted = true;
+        // remaining local contestants stay 'queued' until this one finishes
+      }
+    }
+
+    return { battleId };
+  }
+
+  async function dispatchContestant(
+    battleId: string,
+    projectId: string,
+    battleType: BattleType,
+    prompt: string,
+    c: { id: string; identity: string; model: string; lane: ContestantLane },
+  ): Promise<void> {
+    const { taskId, sessionId } = await dispatch({
+      projectId,
+      contestantId: c.id,
+      prompt,
+      identity: c.identity,
+      model: c.model,
+      battleType,
+    });
+    await sql`
+      UPDATE contestants
+      SET task_id = ${taskId}, status = 'running', updated_at = clock_timestamp()
+      WHERE id = ${c.id}
+    `;
+    publishContestantFrame(battleId, c.id, { status: 'running' });
+    // Start the delta bridge in the background; unsubscribe when the contestant
+    // terminates (teardownDeltaBridge called in handleTaskTerminal).
+    void setupDeltaBridge(battleId, c.id, taskId, sessionId ?? null);
+  }
+
+  // ─── local-lane advance (serialized per battle) ───────────────────────────
+
+  function advanceLocalLane(battleId: string): Promise<void> {
+    const prev = advanceChain.get(battleId) ?? Promise.resolve();
+    const next = prev
+      .catch(() => {})
+      .then(() =>
+        advanceLocalLaneInner(battleId).catch((err) => {
+          log.error({ err: errMsg(err), battleId }, 'arena-runner: advanceLocalLane failed');
+        }),
+      );
+    advanceChain.set(battleId, next);
+    void next.finally(() => {
+      if (advanceChain.get(battleId) === next) advanceChain.delete(battleId);
+    });
+    return next;
+  }
+
+  async function advanceLocalLaneInner(battleId: string): Promise<void> {
+    const battle = await loadBattle(battleId);
+    if (!battle || battle.status !== 'running') return;
+
+    const contestants = await loadContestants(battleId);
+    const slots: ContestantSlot[] = contestants.map((c) => ({
+      id: c.id,
+      lane: c.lane,
+      status: c.status,
+    }));
+
+    // Nothing to do if the local lane is still busy.
+    const localRunning = slots.some((c) => c.lane === 'local' && c.status === 'running');
+    if (localRunning) return;
+
+    const nextId = nextLocalContestant(slots);
+    if (!nextId) return; // local queue is exhausted
+
+    const next = contestants.find((c) => c.id === nextId)!;
+    await dispatchContestant(battleId, battle.project_id, battle.battle_type, battle.prompt, {
+      id: next.id,
+      identity: next.identity,
+      model: next.model,
+      lane: next.lane,
+    });
+  }
+
+  // ─── handleTaskTerminal ───────────────────────────────────────────────────
+
+  function handleTaskTerminal(taskId: string, state: string): void {
+    void (async () => {
+      // Look up which contestant owns this task (contestants_task_id_idx).
+      const [row] = await sql<ContestantRow[]>`
+        SELECT id, battle_id, identity, model, lane, task_id, worktree_id, status
+        FROM contestants WHERE task_id = ${taskId}
+      `;
+      if (!row) return; // not an arena task — ignore
+      if (row.status !== 'running') return; // already settled (idempotent)
+
+      const battle = await loadBattle(row.battle_id);
+
+      // Pull the task row for benchmark + output.
+      const [task] = await sql<{
+        chat_id: string | null;
+        started_at: Date | null;
+        ended_at: Date | null;
+        cost_tokens: number | null;
+      }[]>`SELECT chat_id, started_at, ended_at, cost_tokens FROM tasks WHERE id = ${taskId}`;
+
+      const endedAt = task?.ended_at ?? new Date();
+
+      if (state === 'completed') {
+        const startedAt = task?.started_at ?? endedAt;
+        const bench = computeBenchmark(startedAt, endedAt, task?.cost_tokens ?? null, row.lane);
+
+        const output = task?.chat_id ? await readChatOutput(task.chat_id) : '';
+
+        const resultPath = battle
+          ? await writeContestantResults(battle, row, output, bench).catch((err) => {
+              log.warn({ err: errMsg(err), contestantId: row.id }, 'arena-runner: result write failed');
+              return null;
+            })
+          : null;
+
+        await sql`
+          UPDATE contestants
+          SET status      = 'done',
+              duration_ms = ${Math.round(bench.durationMs)},
+              tokens_per_sec = ${bench.tokensPerSec},
+              cost_tokens = ${task?.cost_tokens ?? null},
+              result_path = ${resultPath},
+              updated_at  = clock_timestamp()
+          WHERE id = ${row.id} AND status = 'running'
+        `;
+        teardownDeltaBridge(row.id);
+
+        // Check if this was the last contestant.
+        const allContestants = await loadContestants(row.battle_id);
+        const battleDone = isBattleComplete(allContestants);
+
+        publishContestantFrame(row.battle_id, row.id, {
+          status: 'done',
+          duration_ms: Math.round(bench.durationMs),
+          ...(bench.tokensPerSec !== null ? { tokens_per_sec: bench.tokensPerSec } : {}),
+          ...(battleDone ? { battle_status: 'completed' } : {}),
+        });
+
+        if (battleDone) {
+          await completeBattle(row.battle_id);
+        } else if (row.lane === 'local') {
+          void advanceLocalLane(row.battle_id);
+        }
+      } else {
+        // failed or cancelled — the contest continues; this contestant is error.
+        const errorMsg = state === 'cancelled' ? 'cancelled' : `task ${state}`;
+        await sql`
+          UPDATE contestants
+          SET status = 'error', error = ${errorMsg}, updated_at = clock_timestamp()
+          WHERE id = ${row.id} AND status = 'running'
+        `;
+        teardownDeltaBridge(row.id);
+
+        const allContestants = await loadContestants(row.battle_id);
+        const battleDone = isBattleComplete(allContestants);
+
+        publishContestantFrame(row.battle_id, row.id, {
+          status: 'error',
+          error: errorMsg,
+          ...(battleDone ? { battle_status: 'completed' } : {}),
+        });
+
+        if (battleDone) {
+          await completeBattle(row.battle_id);
+        } else if (row.lane === 'local') {
+          void advanceLocalLane(row.battle_id);
+        }
+      }
+    })().catch((err) => {
+      log.error({ err: errMsg(err), taskId }, 'arena-runner: handleTaskTerminal failed');
+    });
+  }
+
+  // ─── battle finalization ──────────────────────────────────────────────────
+
+  async function completeBattle(battleId: string): Promise<void> {
+    const updated = await sql`
+      UPDATE battles SET status = 'completed', updated_at = clock_timestamp()
+      WHERE id = ${battleId} AND status = 'running'
+    `;
+    if (updated.count === 0) return; // already terminal (race guard)
+    log.info({ battleId }, 'arena-runner: battle completed');
+
+    // Update manifest with finished_at timestamp.
+    const completedBattle = await loadBattle(battleId);
+    if (completedBattle?.results_path) {
+      const contestants = await loadContestants(battleId);
+      await writeManifest(
+        battleId,
+        completedBattle.results_path,
+        completedBattle.battle_type,
+        completedBattle.prompt,
+        completedBattle.created_at,
+        contestants.map((c) => ({ identity: c.identity, model: c.model, lane: c.lane })),
+        new Date(),
+      ).catch((err) => {
+        log.warn({ err: errMsg(err), battleId }, 'arena-runner: manifest update failed');
+      });
+    }
+
+    onBattleComplete(battleId);
+  }
+
+  // ─── manifest writer ─────────────────────────────────────────────────────
+
+  async function writeManifest(
+    battleId: string,
+    resultsPath: string,
+    battleType: BattleType,
+    prompt: string,
+    createdAt: Date,
+    contestants: Array<{ identity: string; model: string; lane: ContestantLane }>,
+    finishedAt: Date | null,
+  ): Promise<void> {
+    await mkdir(resultsPath, { recursive: true });
+    const manifest = {
+      id: battleId,
+      battle_type: battleType,
+      prompt,
+      contestants,
+      created_at: createdAt.toISOString(),
+      finished_at: finishedAt?.toISOString() ?? null,
+    };
+    await writeFile(join(resultsPath, 'manifest.json'), JSON.stringify(manifest, null, 2), 'utf8');
+  }
+
+  // ─── results writer ───────────────────────────────────────────────────────
+
+  async function writeContestantResults(
+    battle: BattleRow,
+    contestant: { identity: string; model: string; lane: ContestantLane; worktree_id: string | null },
+    output: string,
+    bench: { durationMs: number; tokensPerSec: number | null },
+  ): Promise<string> {
+    const resultsPath = await getOrBuildResultsPath(battle);
+    if (!resultsPath) throw new Error('cannot resolve results path for battle ' + battle.id);
+
+    const contestantDir = buildContestantDir(contestant.identity, contestant.model);
+    const dir = join(resultsPath, contestantDir);
+    await mkdir(dir, { recursive: true });
+
+    const benchLines = [
+      `duration: ${bench.durationMs}ms`,
+      bench.tokensPerSec != null ? `tokens/sec: ${bench.tokensPerSec.toFixed(1)}` : null,
+    ]
+      .filter(Boolean)
+      .join('\n');
+
+    const resultMd =
+      `# ${contestant.identity} / ${contestant.model}\n\n` +
+      `## Benchmark\n\n${benchLines}\n\n` +
+      `## Output\n\n${output}\n`;
+    await writeFile(join(dir, 'result.md'), resultMd, 'utf8');
+
+    if (battle.battle_type === 'coding' && contestant.worktree_id) {
+      const [wt] = await sql<{ path: string; base_commit: string | null }[]>`
+        SELECT path, base_commit FROM worktrees WHERE id = ${contestant.worktree_id}
+      `;
+      if (wt) {
+        const [proj] = await sql<{ path: string }[]>`
+          SELECT path FROM projects WHERE id = ${battle.project_id}
+        `;
+        if (proj) {
+          const diff = await diffWorktree(wt.path, proj.path, {
+            baseRef: wt.base_commit ?? undefined,
+          }).catch(() => '');
+          await writeFile(join(dir, 'diff.patch'), diff, 'utf8');
+        }
+      }
+    }
+
+    return dir;
+  }
+
+  /** Resolve or rebuild results_path for a battle (handles crash-before-UPDATE). */
+  async function getOrBuildResultsPath(battle: BattleRow): Promise<string | null> {
+    if (battle.results_path) return battle.results_path;
+    const [proj] = await sql<{ path: string }[]>`SELECT path FROM projects WHERE id = ${battle.project_id}`;
+    if (!proj) return null;
+    const slug = buildBattleSlug(battle.id, battle.battle_type, battle.created_at);
+    const resultsPath = join(proj.path, 'Arena', slug);
+    await sql`
+      UPDATE battles SET results_path = ${resultsPath}, updated_at = clock_timestamp()
+      WHERE id = ${battle.id}
+    `;
+    return resultsPath;
+  }
+
+  // ─── helpers ──────────────────────────────────────────────────────────────
+
+  async function readChatOutput(chatId: string): Promise<string> {
+    const [m] = await sql<{ content: string | null }[]>`
+      SELECT content FROM messages
+      WHERE chat_id = ${chatId} AND role = 'assistant'
+      ORDER BY created_at DESC LIMIT 1
+    `;
+    return m?.content ?? '';
+  }
+
+  async function loadBattle(battleId: string): Promise<BattleRow | null> {
+    const [b] = await sql<BattleRow[]>`
+      SELECT id, project_id, battle_type, prompt, status, results_path, created_at
+      FROM battles WHERE id = ${battleId}
+    `;
+    return b ?? null;
+  }
+
+  async function loadContestants(battleId: string): Promise<ContestantRow[]> {
+    return sql<ContestantRow[]>`
+      SELECT id, battle_id, identity, model, lane, task_id, worktree_id, status
+      FROM contestants WHERE battle_id = ${battleId}
+      ORDER BY created_at ASC
+    `;
+  }
+
+  function publishContestantFrame(
+    battleId: string,
+    contestantId: string,
+    extra: Record<string, unknown>,
+  ): void {
+    publishUser({
+      type: 'contestant_updated',
+      battle_id: battleId,
+      contestant_id: contestantId,
+      ...extra,
+    });
+  }
+
+  // ─── initResume ───────────────────────────────────────────────────────────
+
+  async function initResume(): Promise<void> {
+    const battles = await sql<BattleRow[]>`
+      SELECT id, project_id, battle_type, prompt, status, results_path, created_at
+      FROM battles WHERE status = 'running'
+    `;
+    if (battles.length === 0) return;
+    log.info({ count: battles.length }, 'arena-runner: resuming in-flight battles on startup');
+    for (const battle of battles) {
+      await resumeBattle(battle).catch((err) => {
+        log.error({ err: errMsg(err), battleId: battle.id }, 'arena-runner: initResume failed for battle');
+      });
+    }
+  }
+
+  async function resumeBattle(battle: BattleRow): Promise<void> {
+    const contestants = await loadContestants(battle.id);
+
+    const taskIds = contestants.map((c) => c.task_id).filter((id): id is string => id !== null);
+    const taskStates = new Map<string, string>();
+    if (taskIds.length > 0) {
+      const tasks = await sql<{ id: string; state: string }[]>`
+        SELECT id, state FROM tasks WHERE id = ANY(${taskIds})
+      `;
+      for (const t of tasks) taskStates.set(t.id, t.state);
+    }
+
+    const decisions = reconcileContestants(
+      contestants.map((c) => ({ contestantId: c.id, taskId: c.task_id, status: c.status })),
+      taskStates,
+    );
+
+    for (const decision of decisions) {
+      if (decision.action === 'keep') continue;
+      const contestant = contestants.find((c) => c.id === decision.contestantId)!;
+      await applyResumeDecision(battle, contestant, decision.action);
+    }
+
+    // Re-check completion after applying decisions.
+    const updated = await loadContestants(battle.id);
+    if (isBattleComplete(updated)) {
+      await completeBattle(battle.id);
+    } else {
+      // Advance local lane in case a slot opened up.
+      void advanceLocalLane(battle.id);
+    }
+
+    log.info({ battleId: battle.id }, 'arena-runner: battle resumed');
+  }
+
+  async function applyResumeDecision(
+    battle: BattleRow,
+    contestant: ContestantRow,
+    action: ContestantResumeAction,
+  ): Promise<void> {
+    switch (action) {
+      case 'keep': break;
+
+      case 'mark-done': {
+        const taskRow = contestant.task_id
+          ? (await sql<{ started_at: Date | null; ended_at: Date | null; cost_tokens: number | null; chat_id: string | null }[]>`
+              SELECT started_at, ended_at, cost_tokens, chat_id FROM tasks WHERE id = ${contestant.task_id}`)[0]
+          : null;
+        const endedAt = taskRow?.ended_at ?? new Date();
+        const startedAt = taskRow?.started_at ?? endedAt;
+        const bench = computeBenchmark(startedAt, endedAt, taskRow?.cost_tokens ?? null, contestant.lane);
+        const output = taskRow?.chat_id ? await readChatOutput(taskRow.chat_id) : '';
+        const resultPath = battle
+          ? await writeContestantResults(battle, contestant, output, bench).catch((err) => {
+              log.warn({ err: errMsg(err), contestantId: contestant.id }, 'arena-runner: resume result write failed');
+              return null;
+            })
+          : null;
+        await sql`
+          UPDATE contestants
+          SET status = 'done',
+              duration_ms = ${Math.round(bench.durationMs)},
+              tokens_per_sec = ${bench.tokensPerSec},
+              result_path = ${resultPath},
+              updated_at = clock_timestamp()
+          WHERE id = ${contestant.id}
+        `;
+        break;
+      }
+
+      case 'mark-error':
+        await sql`
+          UPDATE contestants
+          SET status = 'error', error = 'task failed before callback',
+              updated_at = clock_timestamp()
+          WHERE id = ${contestant.id}
+        `;
+        break;
+
+      case 'mark-cancelled':
+        await sql`
+          UPDATE contestants
+          SET status = 'error', error = 'cancelled before callback',
+              updated_at = clock_timestamp()
+          WHERE id = ${contestant.id}
+        `;
+        break;
+
+      case 're-dispatch': {
+        const { taskId } = await dispatch({
+          projectId: battle.project_id,
+          contestantId: contestant.id,
+          prompt: battle.prompt,
+          identity: contestant.identity,
+          model: contestant.model,
+          battleType: battle.battle_type,
+        });
+        await sql`
+          UPDATE contestants
+          SET task_id = ${taskId}, updated_at = clock_timestamp()
+          WHERE id = ${contestant.id}
+        `;
+        log.info(
+          { battleId: battle.id, contestantId: contestant.id, taskId },
+          'arena-runner: contestant re-dispatched on resume',
+        );
+        break;
+      }
+    }
+  }
+
+  // ─── cancelBattle ─────────────────────────────────────────────────────────
+
+  async function cancelBattle(battleId: string): Promise<{ cancelled: boolean; taskIds: string[] }> {
+    const updated = await sql`
+      UPDATE battles SET status = 'cancelled', updated_at = clock_timestamp()
+      WHERE id = ${battleId} AND status = 'running'
+    `;
+    if (updated.count === 0) return { cancelled: false, taskIds: [] };
+
+    // Mark all non-terminal contestants cancelled and collect in-flight task_ids.
+    const contestants = await sql<{ id: string; task_id: string | null; status: string }[]>`
+      SELECT id, task_id, status FROM contestants
+      WHERE battle_id = ${battleId} AND status NOT IN ('done', 'error')
+    `;
+
+    if (contestants.length > 0) {
+      await sql`
+        UPDATE contestants
+        SET status = 'error', error = 'battle cancelled', updated_at = clock_timestamp()
+        WHERE battle_id = ${battleId} AND status NOT IN ('done', 'error')
+      `;
+      for (const c of contestants) {
+        publishContestantFrame(battleId, c.id, {
+          status: 'error',
+          error: 'battle cancelled',
+          battle_status: 'cancelled',
+        });
+      }
+    }
+
+    const taskIds = contestants
+      .filter(
+        (c): c is typeof c & { task_id: string } =>
+          c.task_id !== null && c.status === 'running',
+      )
+      .map((c) => c.task_id);
+
+    log.info({ battleId }, 'arena-runner: battle cancelled by request');
+    return { cancelled: true, taskIds };
+  }
+
+  // ─── triggerAnalysis (Phase 5 seam) ──────────────────────────────────────
+
+  async function triggerAnalysis(battleId: string): Promise<{ triggered: boolean }> {
+    const battle = await loadBattle(battleId);
+    if (!battle) return { triggered: false };
+    log.info({ battleId }, 'arena-runner: triggerAnalysis requested');
+    // Calls the injected onBattleComplete seam — Phase 5 replaces this with the
+    // real two-stage digest→judge analyzer (see ADR 0002 + plan Phase 5).
+    onBattleComplete(battleId);
+    return { triggered: true };
+  }
+
+  // ─── startCrossExam (Phase 5 seam) ───────────────────────────────────────
+
+  async function startCrossExam(
+    battleId: string,
+    opts: { identity: string; model: string },
+  ): Promise<{ crossExamId: string }> {
+    const [row] = await sql<{ id: string }[]>`
+      INSERT INTO cross_examinations (battle_id, identity, model)
+      VALUES (${battleId}, ${opts.identity}, ${opts.model})
+      RETURNING id
+    `;
+    const crossExamId = row!.id;
+    log.info({ battleId, crossExamId, ...opts }, 'arena-runner: cross-exam inserted, triggering analyzer');
+    if (onCrossExamStart) {
+      try {
+        onCrossExamStart({ battleId, crossExamId, identity: opts.identity, model: opts.model });
+      } catch (err) {
+        log.error({ err: err instanceof Error ? err.message : String(err), battleId, crossExamId }, 'arena-runner: onCrossExamStart threw');
+      }
+    }
+    return { crossExamId };
+  }
+
+  // ─── setWinner (user override) ────────────────────────────────────────────
+
+  async function setWinner(
+    battleId: string,
+    winnerId: string | null,
+  ): Promise<{ ok: boolean; notFound?: boolean; invalidContestant?: boolean }> {
+    const [row] = await sql<{ id: string }[]>`SELECT id FROM battles WHERE id = ${battleId}`;
+    if (!row) return { ok: false, notFound: true };
+
+    if (winnerId !== null) {
+      const [c] = await sql<{ id: string }[]>`
+        SELECT id FROM contestants WHERE id = ${winnerId} AND battle_id = ${battleId}
+      `;
+      if (!c) return { ok: false, invalidContestant: true };
+    }
+
+    await sql`
+      UPDATE battles SET winner_contestant_id = ${winnerId}, updated_at = clock_timestamp()
+      WHERE id = ${battleId}
+    `;
+    publishUser({ type: 'battle_updated', battle_id: battleId, winner_contestant_id: winnerId });
+    return { ok: true };
+  }
+
+  return { startBattle, handleTaskTerminal, initResume, cancelBattle, triggerAnalysis, startCrossExam, setWinner };
+}
+
+function errMsg(e: unknown): string {
+  return e instanceof Error ? e.message : String(e);
+}

apps/coder/src/services/claude-command-discovery.ts

@@ -12,12 +12,48 @@ import { homedir } from 'node:os';
 import { join } from 'node:path';
 import type { AgentCommand } from './provider-types.js';
 
-/** Minimal frontmatter reader — single-line `key: value` between `---` fences. */
+/**
+ * Frontmatter reader between `---` fences. Handles single-line `key: value`
+ * AND YAML block scalars (`key: >` folded / `key: |` literal) whose value
+ * spans the following more-indented lines — the shape most plugin SKILL.md
+ * descriptions use (`description: >`).
+ */
 function frontmatterField(content: string, field: string): string | undefined {
   const block = content.match(/^---\r?\n([\s\S]*?)\r?\n---/);
   if (!block?.[1]) return undefined;
-  const m = block[1].match(new RegExp(`^${field}:\\s*(.+)$`, 'm'));
+  const lines = block[1].split(/\r?\n/);
-  return m?.[1]?.trim().replace(/^["']|["']$/g, '') || undefined;
+  const keyRe = new RegExp(`^(\\s*)${field}:\\s*(.*)$`);
+  for (let i = 0; i < lines.length; i++) {
+    const m = lines[i]?.match(keyRe);
+    if (!m) continue;
+    const keyIndent = (m[1] ?? '').length;
+    const inline = (m[2] ?? '').trim();
+    // Block scalar: `>` (folded) or `|` (literal), optional chomping `+`/`-`.
+    if (/^[>|][+-]?$/.test(inline)) {
+      const folded = inline[0] === '>';
+      const body: string[] = [];
+      for (let j = i + 1; j < lines.length; j++) {
+        const line = lines[j] ?? '';
+        if (line.trim() === '') {
+          body.push('');
+          continue;
+        }
+        const indent = line.length - line.trimStart().length;
+        if (indent <= keyIndent) break; // dedent ends the block
+        body.push(line.slice(keyIndent + 1));
+      }
+      const joined = folded
+        ? body
+            .map((l) => l.trim())
+            .join(' ')
+            .replace(/\s+/g, ' ')
+            .trim()
+        : body.join('\n').replace(/\n+$/, '');
+      return joined || undefined;
+    }
+    return inline.replace(/^["']|["']$/g, '').trim() || undefined;
+  }
+  return undefined;
 }
 
 function readCommandDir(dir: string): AgentCommand[] {

apps/coder/src/services/dispatcher.ts

@@ -4,6 +4,7 @@ import type { Broker } from '@boocode/server/broker';
 import type { WsFrame } from '@boocode/contracts/ws-frames';
 import type { Config } from '../config.js';
 import { createWorktree, diffWorktree, cleanupWorktree, ensureSessionWorktree } from './worktrees.js';
+import { asPermissionMode } from './tools/types.js';
 import { createCheckpoint } from './checkpoints.js';
 import { makeDcpStreamStripper } from './dcp-strip.js';
 import { dispatchViaAcp } from './acp-dispatch.js';
@@ -31,7 +32,13 @@ import {
 import { shouldFailOnMissingAgent } from './flow-runner-decisions.js';
 
 interface InferenceRunner {
-  enqueue: (sessionId: string, chatId: string, assistantId: string, user: string) => void;
+  enqueue: (
+    sessionId: string,
+    chatId: string,
+    assistantId: string,
+    user: string,
+    permissionMode?: 'plan' | 'ask' | 'bypass',
+  ) => void;
   cancel: (sessionId: string, chatId: string) => Promise<boolean>;
   hasActive: (chatId: string) => boolean;
 }
@@ -305,10 +312,13 @@ export function createDispatcher(deps: Deps): {
 
   // ─── Path A: Native Inference ───────────────────────────────────────────────
 
-  async function runNativeInference(task: { id: string; project_id: string; input: string; agent: string | null; model: string | null; session_id: string | null }): Promise<void> {
+  async function runNativeInference(task: { id: string; project_id: string; input: string; agent: string | null; model: string | null; mode_id: string | null; session_id: string | null }): Promise<void> {
     const taskId = task.id;
     log.info({ taskId }, 'dispatcher: starting task (path A — native)');
 
+    // Declared before try so the catch block can write it back on the task row.
+    let chatId: string | null = null;
+
     try {
       // Mark running
       await sql`
@@ -317,26 +327,29 @@ export function createDispatcher(deps: Deps): {
         WHERE id = ${taskId}
       `;
 
-      // Create session + chat for this task
+      // Session setup: reuse a pre-created session (e.g. Q&A arena contestants
+      // whose persona is stamped on the session via agent_id) or create a fresh one.
       const model = task.model ?? config.DEFAULT_MODEL;
-      const sessionName = 'Task: ' + task.input.slice(0, 40);
+      let sessionId: string;
-
+      if (task.session_id) {
-      const [session] = await sql<{ id: string }[]>`
+        sessionId = task.session_id;
-        INSERT INTO sessions (project_id, name, model, status)
+      } else {
-        VALUES (${task.project_id}, ${sessionName}, ${model}, 'open')
+        const sessionName = 'Task: ' + task.input.slice(0, 40);
-        RETURNING id
+        const [session] = await sql<{ id: string }[]>`
-      `;
+          INSERT INTO sessions (project_id, name, model, status)
-      const sessionId = session!.id;
+          VALUES (${task.project_id}, ${sessionName}, ${model}, 'open')
+          RETURNING id
+        `;
+        sessionId = session!.id;
+        await sql`UPDATE tasks SET session_id = ${sessionId} WHERE id = ${taskId}`;
+      }
 
       const [chat] = await sql<{ id: string }[]>`
         INSERT INTO chats (session_id, name, status)
         VALUES (${sessionId}, 'Task execution', 'open')
         RETURNING id
       `;
-      const chatId = chat!.id;
+      chatId = chat!.id;
-
-      // Link task to session
-      await sql`UPDATE tasks SET session_id = ${sessionId} WHERE id = ${taskId}`;
 
       // Create user message + streaming assistant
       await sql<{ id: string }[]>`
@@ -351,8 +364,9 @@ export function createDispatcher(deps: Deps): {
       `;
       const assistantId = assistantMsg!.id;
 
-      // Enqueue inference
+      // Enqueue inference — pass the native permission gate (plan/ask/bypass)
-      inference.enqueue(sessionId, chatId, assistantId, 'default');
+      // through to the write-tool context. Non-unified mode ids → undefined.
+      inference.enqueue(sessionId, chatId, assistantId, 'default', asPermissionMode(task.mode_id));
 
       // Wait for inference to complete (poll message status)
       const finalStatus = await waitForCompletion(assistantId);
@@ -381,7 +395,7 @@ export function createDispatcher(deps: Deps): {
         const summary = (msg?.content ?? '').slice(0, 500);
         await sql`
           UPDATE tasks
-          SET state = 'completed', ended_at = clock_timestamp(), output_summary = ${summary}, cost_tokens = ${costTokens}
+          SET state = 'completed', ended_at = clock_timestamp(), output_summary = ${summary}, cost_tokens = ${costTokens}, chat_id = ${chatId}
           WHERE id = ${taskId}
         `;
         log.info({ taskId, costTokens }, 'dispatcher: task completed (native)');
@@ -392,7 +406,7 @@ export function createDispatcher(deps: Deps): {
         const summary = (msg?.content ?? 'Inference failed').slice(0, 500);
         await sql`
           UPDATE tasks
-          SET state = 'failed', ended_at = clock_timestamp(), output_summary = ${summary}, cost_tokens = ${costTokens}
+          SET state = 'failed', ended_at = clock_timestamp(), output_summary = ${summary}, cost_tokens = ${costTokens}, chat_id = ${chatId}
           WHERE id = ${taskId}
         `;
         log.warn({ taskId, finalStatus }, 'dispatcher: task failed (native)');
@@ -402,7 +416,7 @@ export function createDispatcher(deps: Deps): {
       log.error({ taskId, err: errMsg }, 'dispatcher: task error (native)');
       await sql`
         UPDATE tasks
-        SET state = 'failed', ended_at = clock_timestamp(), output_summary = ${errMsg.slice(0, 500)}
+        SET state = 'failed', ended_at = clock_timestamp(), output_summary = ${errMsg.slice(0, 500)}, chat_id = ${chatId}
         WHERE id = ${taskId}
       `.catch(() => {});
     }

apps/coder/src/services/edit-guards-imports.ts

@@ -0,0 +1,47 @@
+// edit-guards-imports — detects dropped imports in edited files.
+// Ported from opencode-morph-fast-apply (MIT).
+
+export interface ImportCheckResult {
+  ok: boolean;
+  missingImports: string[];
+  reason?: string;
+}
+
+const IMPORT_PATTERNS = [
+  /^import\s+(?:\{[^}]*\}|\*\s+as\s+\w+|\w+)\s+from\s+['"][^'"]+['"]\s*;?$/m,
+  /^import\s+['"][^'"]+['"]\s*;?$/m,
+  /^export\s+.*\s+from\s+['"][^'"]+['"]\s*;?$/m,
+  /^require\s*\(\s*['"][^'"]+['"]\s*\)\s*;?$/m,
+  /^import\s+type\s+\{[^}]*\}\s+from\s+['"][^'"]+['"]\s*;?$/m,
+];
+
+function extractImportLines(content: string): string[] {
+  return content.split('\n').filter((line) =>
+    IMPORT_PATTERNS.some((p) => p.test(line.trim())),
+  );
+}
+
+export function checkDroppedImports(
+  original: string,
+  updated: string,
+  filePath: string,
+): ImportCheckResult {
+  const originalImports = extractImportLines(original);
+  const updatedImports = extractImportLines(updated);
+
+  if (originalImports.length === 0) {
+    return { ok: true, missingImports: [] };
+  }
+
+  const missing = originalImports.filter((imp) => !updatedImports.includes(imp));
+
+  if (missing.length > 0 && originalImports.length > 0) {
+    return {
+      ok: false,
+      missingImports: missing,
+      reason: `Edit would drop ${missing.length} import(s) from ${filePath}`,
+    };
+  }
+
+  return { ok: true, missingImports: [] };
+}

apps/coder/src/services/edit-guards.ts

@@ -0,0 +1,42 @@
+// v2.8 Morph safety guards — prevents catastrophic truncation, marker leakage,
+// and accidental import deletion during native edit_file application.
+// Ported from opencode-morph-fast-apply (MIT) with threshold values preserved.
+
+export interface GuardResult {
+  ok: boolean;
+  reason?: string;
+  charLoss?: number;
+  lineLoss?: number;
+}
+
+const TRUNCATION_CHAR_THRESHOLD = 0.6;
+const TRUNCATION_LINE_THRESHOLD = 0.5;
+
+export function validateEditResult(
+  original: string,
+  updated: string,
+  filePath: string,
+): GuardResult {
+  // Check for catastrophic content truncation
+  if (original.length > 0 && updated.length > 0) {
+    const charLoss = 1 - updated.length / original.length;
+    const originalLines = original.split('\n').length;
+    const updatedLines = updated.split('\n').length;
+    const lineLoss = 1 - updatedLines / originalLines;
+
+    if (charLoss > TRUNCATION_CHAR_THRESHOLD && lineLoss > TRUNCATION_LINE_THRESHOLD) {
+      return {
+        ok: false,
+        reason: `Edit would truncate ${Math.round(charLoss * 100)}% of characters and ${Math.round(lineLoss * 100)}% of lines`,
+        charLoss,
+        lineLoss,
+      };
+    }
+  }
+
+  return { ok: true };
+}
+
+export function formatGuardError(guard: GuardResult, filePath: string): string {
+  return `Edit guard rejected change to ${filePath}: ${guard.reason ?? 'unknown error'}`;
+}

apps/coder/src/services/flow-artifacts.ts

@@ -0,0 +1,23 @@
+import { mkdir, writeFile } from 'node:fs/promises';
+import { join } from 'node:path';
+import { existsSync } from 'node:fs';
+
+const ARTIFACTS_ROOT = 'data/flow-artifacts';
+
+export function getArtifactPath(flowRunId: string, stepId: string): string {
+  return join(ARTIFACTS_ROOT, flowRunId, `${stepId}.md`);
+}
+
+export async function writeFlowArtifact(
+  flowRunId: string,
+  stepId: string,
+  content: string,
+): Promise<string> {
+  const dir = join(ARTIFACTS_ROOT, flowRunId);
+  if (!existsSync(dir)) {
+    await mkdir(dir, { recursive: true });
+  }
+  const path = getArtifactPath(flowRunId, stepId);
+  await writeFile(path, content, 'utf8');
+  return path;
+}

apps/coder/src/services/flow-runner-decisions.ts

@@ -22,7 +22,7 @@
  * "Settled" = done ∪ skipped ∪ excluded. Only settled deps unblock a step;
  * an inFlight dep does NOT (the runner waits for its terminal callback).
  */
-import type { Flow, Step, StepContext } from '../conductor/types.js';
+import type { Flow, Step, StepContext, TriggerRule } from '../conductor/types.js';
 
 export interface SchedulerState {
   /** step ids that completed successfully (results available) */
@@ -62,7 +62,7 @@ export function readySteps(flow: Flow, state: SchedulerState): Step[] {
       !state.skipped.has(s.id) &&
       !state.inFlight.has(s.id) &&
       !state.excluded.has(s.id) &&
-      (s.deps ?? []).every((d) => isSatisfied(state, d)),
+      ((s.deps ?? []).length === 0 || evaluateTriggerRule(s.deps ?? [], state.done, state.skipped, state.excluded, s.trigger_rule)),
   );
 }
 
@@ -167,6 +167,32 @@ export function shouldFailOnMissingAgent(agent: string, modeId: string | null):
   return agent === 'qwen' && modeId === 'plan';
 }
 
+/**
+ * Evaluate a trigger rule against dependency results.
+ * - all_success: every dep must be done (not skipped/failed)
+ * - one_success: at least one dep must be done
+ * - all_done: every dep must be settled regardless of outcome
+ */
+export function evaluateTriggerRule(
+  deps: string[],
+  done: ReadonlySet<string>,
+  skipped: ReadonlySet<string>,
+  excluded: ReadonlySet<string>,
+  rule?: TriggerRule,
+): boolean {
+  if (deps.length === 0) return true;
+  const satisfied = new Set([...done, ...skipped, ...excluded]);
+
+  switch (rule ?? 'all_success') {
+    case 'all_success':
+      return deps.every((d) => done.has(d) || skipped.has(d) || excluded.has(d));
+    case 'one_success':
+      return deps.some((d) => done.has(d));
+    case 'all_done':
+      return deps.every((d) => satisfied.has(d));
+  }
+}
+
 /**
  * Reconcile every step of an in-flight run for startup resume. Returns one
  * decision per step. Pure — no IO.

apps/coder/src/services/flow-runner.ts

@@ -346,6 +346,20 @@ export function createFlowRunner(deps: Deps): FlowRunner {
         continue; // re-evaluate — code output can unblock the next wave
       }
 
+      // Approval gate steps: pause and wait for human decision.
+      const approvalReady = toRun.filter((s) => s.kind === 'approval');
+      if (approvalReady.length > 0) {
+        for (const s of approvalReady) {
+          await sql`
+            UPDATE flow_steps SET status = 'blocked', updated_at = clock_timestamp()
+            WHERE run_id = ${runId} AND step_id = ${s.id}
+          `;
+          await appendStepEvent(sql, runId, s.id, 'paused', { reason: 'awaiting approval' });
+          publishStep(runId, s.id, 'blocked');
+        }
+        return;
+      }
+
       // Only agent steps remain ready → dispatch the whole parallel wave, then wait.
       for (const s of toRun) {
         await dispatchAgentStep(runId, run.project_id, model, s, ctx);
@@ -378,7 +392,8 @@ export function createFlowRunner(deps: Deps): FlowRunner {
     // flow's step.run already bakes in the evidence/YAGNI contracts.
     const persona = step.agent ? await loadPersona(step.agent) : '';
     const taskPrompt = await step.run(ctx);
-    const fullPrompt = persona ? `${persona}\n\n---\n\n${taskPrompt}` : taskPrompt;
+    const resolvedPrompt = resolveVariables(taskPrompt, ctx.results);
+    const fullPrompt = persona ? `${persona}\n\n---\n\n${resolvedPrompt}` : resolvedPrompt;
 
     // READ-ONLY (D-4): agent='qwen', mode_id='plan' are hardcoded, never
     // user-overridable. The dispatcher's qwen+plan rule forces the PTY hard gate.
@@ -392,6 +407,7 @@ export function createFlowRunner(deps: Deps): FlowRunner {
       SET task_id = ${task!.id}, status = 'running', input = ${fullPrompt}, updated_at = clock_timestamp()
       WHERE run_id = ${runId} AND step_id = ${step.id}
     `;
+    await appendStepEvent(sql, runId, step.id, 'started', { taskId: task!.id });
   }
 
   /**
@@ -438,6 +454,7 @@ export function createFlowRunner(deps: Deps): FlowRunner {
         WHERE run_id = ${runId} AND step_id = ${stepId}
       `;
     }
+    await appendStepEvent(sql, runId, stepId, status, output ? { outputLength: output.length } : undefined);
   }
 
   // ─── run completion ─────────────────────────────────────────────────────────
@@ -483,6 +500,7 @@ export function createFlowRunner(deps: Deps): FlowRunner {
     if (updated.count === 0) return;
     const stepId = failedStepId ?? (flow ? lastAgentStepId(flow, input, model) : 'run');
     log.warn({ runId, error }, 'flow-runner: run failed');
+    await appendStepEvent(sql, runId, stepId, 'failed', { error });
     publishStep(runId, stepId, 'failed', { run_status: 'failed' });
   }
 
@@ -522,7 +540,7 @@ export function createFlowRunner(deps: Deps): FlowRunner {
   function publishStep(
     runId: string,
     stepId: string,
-    status: 'running' | 'completed' | 'failed' | 'skipped' | 'cancelled',
+    status: 'running' | 'completed' | 'failed' | 'skipped' | 'cancelled' | 'blocked',
     extra?: { run_status?: 'running' | 'completed' | 'failed' | 'cancelled'; report?: string },
   ): void {
     publishUser({
@@ -763,3 +781,40 @@ export function createFlowRunner(deps: Deps): FlowRunner {
 function errMsg(e: unknown): string {
   return e instanceof Error ? e.message : String(e);
 }
+
+// ─── Event log ───────────────────────────────────────────────────────────────
+
+async function appendStepEvent(
+  sql: Sql,
+  runId: string,
+  stepId: string,
+  event: string,
+  payload?: Record<string, unknown>,
+): Promise<void> {
+  await sql`
+    INSERT INTO flow_step_events (run_id, step_id, event, payload)
+    VALUES (${runId}, ${stepId}, ${event}, ${payload ? sql.json(payload as never) : null})
+  `;
+}
+
+// ─── Variable substitution ───────────────────────────────────────────────────
+
+const VAR_PATTERN = /\$(\w+)\.output(?:\.(\w+(?:\.\w+)*))?/g;
+
+export function resolveVariables(prompt: string, results: Record<string, string>): string {
+  return prompt.replace(VAR_PATTERN, (match, stepId, fieldPath) => {
+    const output = results[stepId];
+    if (!output) return match;
+    if (!fieldPath) return output;
+    try {
+      const lines = output.split('\n');
+      for (const line of lines) {
+        const parsed = line.match(new RegExp(`^${fieldPath}:\\s*(.+)$`, 'i'));
+        if (parsed) return parsed[1]!.trim();
+      }
+    } catch {
+      // fall through
+    }
+    return match;
+  });
+}

apps/coder/src/services/fuzzy-match.ts

@@ -21,7 +21,16 @@
 //                         punctuation to ASCII on both sides; the match is
 //                         mapped back to original offsets.
 //   4. levenshtein      — best line-window by normalized edit-distance
-//                         similarity; accepted only at >= SIMILARITY_THRESHOLD.
+//                         similarity; accepted only at >= SIMILARITY_THRESHOLD,
+//                         anchored on an exact first+last line for multi-line
+//                         needles, and REFUSED (ambiguous) when a second window
+//                         scores within AMBIGUITY_EPSILON of the best. Like the
+//                         exact/whitespace tiers, this tier fails CLOSED — it
+//                         never splices over a merely-plausible guess, because a
+//                         wrong-window splice corrupts the file (it leaves the
+//                         real target intact and duplicates it). This mirrors
+//                         opencode/cline/qwen, whose fuzzy tiers all keep the
+//                         unique-match requirement rather than picking a winner.
 //
 // Pure and dependency-free (Levenshtein is the standard iterative two-row DP),
 // reimplemented from the general technique — no vendored source.
@@ -31,8 +40,31 @@ export type MatchResult =
   | { kind: 'ambiguous'; count: number }
   | { kind: 'not_found' };
 
-/** Levenshtein similarity floor for the final fuzzy fallback (strategy 4). */
+/**
-export const SIMILARITY_THRESHOLD = 0.66;
+ * Levenshtein similarity floor for the final fuzzy fallback (strategy 4).
+ * 0.66 was far too low — at two-thirds similarity a structurally-wrong window
+ * (e.g. one of three near-identical form blocks) clears the bar and gets spliced
+ * over, leaving the real target intact and duplicated. Competent agents anchor
+ * far tighter (opencode's BlockAnchor needs an exact anchor; cline needs exact
+ * first+last lines). 0.85 keeps genuine quantized-model drift (a typo, an indent
+ * shift) while refusing a different block.
+ */
+export const SIMILARITY_THRESHOLD = 0.85;
+
+/**
+ * If a second candidate window scores within this of the best, the match is
+ * ambiguous and tier 4 refuses rather than guessing — the same fail-closed
+ * stance the exact and whitespace tiers take on multiple hits. Repetitive files
+ * (the duplicate-block corruption case) produce near-tied windows; this is what
+ * turns that into a clean "add more context" error instead of a wrong splice.
+ */
+export const AMBIGUITY_EPSILON = 0.05;
+
+/** Multi-line needles at or above this length must anchor on an exact (after
+ *  trim + unicode-fold) first AND last line before similarity is even scored —
+ *  the cline/opencode block-anchor rule. Below it, threshold + uniqueness alone
+ *  guard the match. */
+const ANCHOR_MIN_LINES = 3;
 
 export function locateMatch(content: string, needle: string): MatchResult {
   // Empty needle has no meaningful match.
@@ -252,20 +284,39 @@ function locateByLevenshtein(content: string, needle: string): MatchResult | nul
 
   const needleJoined = needleLines.map((l) => l.trim()).join('\n');
 
-  let best = -1;
+  // Block-anchor gate for multi-line needles: the first and last lines must match
-  let bestSpan: { start: number; end: number } | null = null;
+  // exactly (after trim + unicode-fold) or the window is not even scored. This
+  // stops a high interior-similarity from dragging a structurally-wrong window
+  // over the threshold — the failure that duplicates blocks in repetitive files.
+  const anchored = n >= ANCHOR_MIN_LINES;
+  const needleFirst = canonicalize(needleLines[0]!.trim());
+  const needleLast = canonicalize(needleLines[n - 1]!.trim());
+
+  const scored: Array<{ score: number; start: number; end: number }> = [];
   for (let i = 0; i + n <= contentLines.length; i++) {
     const window = contentLines.slice(i, i + n);
-    const windowJoined = window.map((l) => l.text.trim()).join('\n');
+    if (anchored) {
-    const score = similarity(windowJoined, needleJoined);
+      const winFirst = canonicalize(window[0]!.text.trim());
-    if (score > best) {
+      const winLast = canonicalize(window[n - 1]!.text.trim());
-      best = score;
+      if (winFirst !== needleFirst || winLast !== needleLast) continue;
-      bestSpan = { start: window[0]!.start, end: window[n - 1]!.end };
     }
+    const windowJoined = window.map((l) => l.text.trim()).join('\n');
+    scored.push({
+      score: similarity(windowJoined, needleJoined),
+      start: window[0]!.start,
+      end: window[n - 1]!.end,
+    });
   }
 
-  if (bestSpan && best >= SIMILARITY_THRESHOLD) {
+  if (scored.length === 0) return null;
-    return { kind: 'fuzzy', start: bestSpan.start, end: bestSpan.end };
+  scored.sort((a, b) => b.score - a.score);
-  }
+  const best = scored[0]!;
-  return null;
+  if (best.score < SIMILARITY_THRESHOLD) return null;
+
+  // Uniqueness guard: refuse when a second window is within epsilon of the best.
+  // Fail closed (ambiguous) rather than silently splicing one of several lookalikes.
+  const tied = scored.filter((s) => s.score >= best.score - AMBIGUITY_EPSILON);
+  if (tied.length > 1) return { kind: 'ambiguous', count: tied.length };
+
+  return { kind: 'fuzzy', start: best.start, end: best.end };
 }

apps/coder/src/services/lsp/client.ts

@@ -0,0 +1,75 @@
+import { createInterface } from 'node:readline';
+import type { Readable, Writable } from 'node:stream';
+
+interface RpcRequest {
+  jsonrpc: '2.0';
+  id: number;
+  method: string;
+  params?: unknown;
+}
+
+interface RpcResponse {
+  jsonrpc: '2.0';
+  id: number;
+  result?: unknown;
+  error?: { code: number; message: string };
+}
+
+export class LspClient {
+  private nextId = 1;
+  private pending = new Map<number, { resolve: (v: RpcResponse) => void; reject: (e: Error) => void }>();
+  private buffer = '';
+
+  constructor(
+    private stdin: Writable,
+    private stdout: Readable,
+  ) {
+    const rl = createInterface({ input: stdout, crlfDelay: Infinity });
+    rl.on('line', (line) => this.handleLine(line));
+  }
+
+  private handleLine(line: string): void {
+    this.buffer += line + '\n';
+    const match = this.buffer.match(/Content-Length: (\d+)\r?\n\r?\n/);
+    if (!match || !match[1]) return;
+    const len = parseInt(match[1], 10);
+    const headerEnd = match.index! + match[0].length;
+    const body = this.buffer.slice(headerEnd, headerEnd + len);
+    if (body.length < len) return;
+    this.buffer = this.buffer.slice(headerEnd + len);
+    try {
+      const msg: RpcResponse = JSON.parse(body);
+      const cb = this.pending.get(msg.id);
+      if (cb) {
+        this.pending.delete(msg.id);
+        cb.resolve(msg);
+      }
+    } catch {
+      // Malformed JSON, ignore
+    }
+  }
+
+  async request(method: string, params?: unknown): Promise<unknown> {
+    const id = this.nextId++;
+    const req: RpcRequest = { jsonrpc: '2.0', id, method, params };
+    const body = JSON.stringify(req);
+    const header = `Content-Length: ${Buffer.byteLength(body, 'utf8')}\r\n\r\n`;
+
+    return new Promise((resolve, reject) => {
+      this.pending.set(id, {
+        resolve: (resp) => {
+          if (resp.error) reject(new Error(resp.error.message));
+          else resolve(resp.result);
+        },
+        reject,
+      });
+      this.stdin.write(header + body);
+    });
+  }
+
+  async notify(method: string, params?: unknown): Promise<void> {
+    const body = JSON.stringify({ jsonrpc: '2.0', method, params });
+    const header = `Content-Length: ${Buffer.byteLength(body, 'utf8')}\r\n\r\n`;
+    this.stdin.write(header + body);
+  }
+}

apps/coder/src/services/lsp/config.ts

@@ -0,0 +1,19 @@
+export interface LspServerConfig {
+  command: string;
+  args: string[];
+  rootPatterns: string[];
+}
+
+const TS_CONFIG: LspServerConfig = {
+  command: 'typescript-language-server',
+  args: ['--stdio'],
+  rootPatterns: ['package.json', 'tsconfig.json'],
+};
+
+const SUPPORTED_EXTS = new Set(['ts', 'tsx', 'js', 'jsx', 'mjs', 'cjs']);
+
+export function getServerConfig(filePath: string): LspServerConfig | null {
+  const ext = filePath.split('.').pop()?.toLowerCase();
+  if (ext && SUPPORTED_EXTS.has(ext)) return TS_CONFIG;
+  return null;
+}

apps/coder/src/services/lsp/operations.ts

@@ -0,0 +1,86 @@
+import type { LspClient } from './client.js';
+import type { Diagnostic, Location } from './types.js';
+
+function fileUri(filePath: string): string {
+  return `file://${filePath.startsWith('/') ? '' : '/'}${filePath}`;
+}
+
+export async function openDocument(
+  client: LspClient,
+  filePath: string,
+  content: string,
+  version: number = 1,
+): Promise<void> {
+  const uri = fileUri(filePath);
+  await client.notify('textDocument/didOpen', {
+    textDocument: { uri, languageId: 'typescript', version, text: content },
+  });
+}
+
+export async function closeDocument(client: LspClient, filePath: string): Promise<void> {
+  await client.notify('textDocument/didClose', {
+    textDocument: { uri: fileUri(filePath) },
+  });
+}
+
+export async function getDiagnostics(
+  client: LspClient,
+  filePath: string,
+  content: string,
+): Promise<Diagnostic[]> {
+  const uri = fileUri(filePath);
+  await openDocument(client, filePath, content);
+  const result: any = await client.request('textDocument/diagnostic', {
+    textDocument: { uri },
+  });
+  await closeDocument(client, filePath);
+  const diagnostics: Diagnostic[] = [];
+  if (result?.diagnostics) {
+    for (const d of result.diagnostics) {
+      diagnostics.push({
+        range: d.range,
+        severity: d.severity ?? 1,
+        message: d.message,
+        source: d.source,
+      });
+    }
+  }
+  return diagnostics;
+}
+
+export async function gotoDefinition(
+  client: LspClient,
+  filePath: string,
+  content: string,
+  line: number,
+  character: number,
+): Promise<Location | null> {
+  const uri = fileUri(filePath);
+  await openDocument(client, filePath, content);
+  const result: any = await client.request('textDocument/definition', {
+    textDocument: { uri },
+    position: { line, character },
+  });
+  await closeDocument(client, filePath);
+  if (!result) return null;
+  const loc = Array.isArray(result) ? result[0] : result;
+  return loc ? { uri: loc.uri, range: loc.range } : null;
+}
+
+export async function findReferences(
+  client: LspClient,
+  filePath: string,
+  content: string,
+  line: number,
+  character: number,
+): Promise<Location[]> {
+  const uri = fileUri(filePath);
+  await openDocument(client, filePath, content);
+  const result: any = await client.request('textDocument/references', {
+    textDocument: { uri },
+    position: { line, character },
+    context: { includeDeclaration: true },
+  });
+  await closeDocument(client, filePath);
+  return (result ?? []).map((loc: any) => ({ uri: loc.uri, range: loc.range }));
+}

apps/coder/src/services/lsp/server-manager.ts

@@ -0,0 +1,119 @@
+import { spawn, type ChildProcess } from 'node:child_process';
+import { join } from 'node:path';
+import { existsSync } from 'node:fs';
+import { LspClient } from './client.js';
+import { getServerConfig } from './config.js';
+
+const IDLE_TIMEOUT_MS = 5 * 60 * 1000;
+const SWEEP_INTERVAL_MS = 30_000;
+
+interface LspInstance {
+  client: LspClient;
+  proc: ChildProcess;
+  lastUsed: number;
+  timer: ReturnType<typeof setTimeout>;
+}
+
+export class LspServerManager {
+  private instances = new Map<string, LspInstance>();
+  private sweepTimer: ReturnType<typeof setInterval> | null = null;
+
+  constructor() {
+    this.startSweeper();
+  }
+
+  private startSweeper(): void {
+    this.sweepTimer = setInterval(() => this.sweep(), SWEEP_INTERVAL_MS);
+    this.sweepTimer.unref?.();
+  }
+
+  private findProjectRoot(filePath: string): string | null {
+    let dir = filePath;
+    const config = getServerConfig(filePath);
+    if (!config) return null;
+    while (true) {
+      for (const pattern of config.rootPatterns) {
+        if (existsSync(join(dir, pattern))) return dir;
+      }
+      const parent = join(dir, '..');
+      if (parent === dir) return dir;
+      dir = parent;
+    }
+  }
+
+  async getClient(filePath: string): Promise<LspClient | null> {
+    const config = getServerConfig(filePath);
+    if (!config) return null;
+    const projectRoot = this.findProjectRoot(filePath);
+    if (!projectRoot) return null;
+
+    const existing = this.instances.get(projectRoot);
+    if (existing) {
+      existing.lastUsed = Date.now();
+      clearTimeout(existing.timer);
+      existing.timer = setTimeout(() => this.kill(projectRoot), IDLE_TIMEOUT_MS);
+      existing.timer.unref?.();
+      return existing.client;
+    }
+
+    return this.spawn(projectRoot, config.command, config.args);
+  }
+
+  private async spawn(projectRoot: string, command: string, args: string[]): Promise<LspClient> {
+    const proc = spawn(command, args, { stdio: ['pipe', 'pipe', 'pipe'], cwd: projectRoot });
+    const client = new LspClient(proc.stdin!, proc.stdout!);
+
+    await client.request('initialize', {
+      processId: process.pid,
+      rootUri: `file://${projectRoot}`,
+      capabilities: {
+        textDocument: {
+          diagnostic: { dynamicRegistration: false },
+          definition: { dynamicRegistration: false },
+          references: { dynamicRegistration: false },
+        },
+      },
+    });
+    await client.notify('initialized', {});
+
+    const timer = setTimeout(() => this.kill(projectRoot), IDLE_TIMEOUT_MS);
+    timer.unref?.();
+
+    this.instances.set(projectRoot, { client, proc, lastUsed: Date.now(), timer });
+    proc.on('exit', () => this.instances.delete(projectRoot));
+
+    return client;
+  }
+
+  private kill(projectRoot: string): void {
+    const inst = this.instances.get(projectRoot);
+    if (!inst) return;
+    this.instances.delete(projectRoot);
+    inst.proc.kill('SIGTERM');
+    setTimeout(() => {
+      if (inst.proc.exitCode === null) inst.proc.kill('SIGKILL');
+    }, 5000);
+  }
+
+  private sweep(): void {
+    const now = Date.now();
+    for (const [root, inst] of this.instances) {
+      if (now - inst.lastUsed > IDLE_TIMEOUT_MS) {
+        this.kill(root);
+      }
+    }
+  }
+
+  shutdown(): void {
+    if (this.sweepTimer) clearInterval(this.sweepTimer);
+    for (const root of [...this.instances.keys()]) {
+      this.kill(root);
+    }
+  }
+
+  getActiveCount(): number {
+    return this.instances.size;
+  }
+}
+
+export const lspManager = new LspServerManager();

apps/coder/src/services/lsp/types.ts

@@ -0,0 +1,28 @@
+export interface Position {
+  line: number;
+  character: number;
+}
+
+export interface Range {
+  start: Position;
+  end: Position;
+}
+
+export interface Location {
+  uri: string;
+  range: Range;
+}
+
+export interface Diagnostic {
+  range: Range;
+  severity: number;
+  message: string;
+  source?: string;
+}
+
+export interface TextDocumentItem {
+  uri: string;
+  languageId: string;
+  version: number;
+  text: string;
+}

apps/coder/src/services/pending_changes.ts

@@ -1,9 +1,120 @@
-import { readFile, writeFile, unlink, mkdir } from 'node:fs/promises';
+import { readFile, writeFile, unlink, mkdir, rename, realpath } from 'node:fs/promises';
-import { dirname } from 'node:path';
+import { dirname, join, basename } from 'node:path';
+import { randomBytes } from 'node:crypto';
 import type { Sql } from '../db.js';
 import { resolveWritePath } from './write_guard.js';
 import { locateMatch } from './fuzzy-match.js';
 
+/**
+ * Write a file atomically: stage to a sibling temp file, then rename over the
+ * target. rename(2) on the same filesystem is atomic, so a crash mid-write can
+ * never leave a half-written (truncated/corrupt) source file — readers see
+ * either the old content or the complete new content. The temp lives in the same
+ * directory to guarantee a same-filesystem rename.
+ *
+ * Symlinks: a plain writeFile FOLLOWS a symlink and writes through to its target;
+ * a bare rename would REPLACE the link with a regular file. We realpath an
+ * existing target first so the rename lands on the real file and the link
+ * survives — preserving the prior follow-through behavior. A missing target
+ * (create, or a broken link) just writes the literal path.
+ */
+async function writeFileAtomic(filePath: string, content: string): Promise<void> {
+  let target = filePath;
+  try {
+    target = await realpath(filePath);
+  } catch {
+    // ENOENT (new file) or broken link — write the literal path.
+  }
+  const tmp = join(dirname(target), `.${basename(target)}.tmp.${process.pid}.${randomBytes(6).toString('hex')}`);
+  await writeFile(tmp, content, 'utf8');
+  try {
+    await rename(tmp, target);
+  } catch (err) {
+    await unlink(tmp).catch(() => {});
+    throw err;
+  }
+}
+
+/** Detect a file's dominant line ending so an edit can preserve it. */
+function detectEol(text: string): '\r\n' | '\n' {
+  return text.includes('\r\n') ? '\r\n' : '\n';
+}
+
+/**
+ * Serialize the read-modify-write of a single file so two concurrent applies
+ * (e.g. two chat tabs sharing one worktree, or a Bypass write racing an
+ * apply_pending) can't lose an update. In-process keying is sufficient —
+ * BooCoder is a single Fastify process. One Map entry per distinct path.
+ */
+const fileLocks = new Map<string, Promise<void>>();
+async function withFileLock<T>(filePath: string, fn: () => Promise<T>): Promise<T> {
+  const prev = fileLocks.get(filePath) ?? Promise.resolve();
+  let release!: () => void;
+  const current = new Promise<void>((r) => { release = r; });
+  fileLocks.set(filePath, prev.then(() => current));
+  await prev.catch(() => {});
+  try {
+    return await fn();
+  } finally {
+    release();
+  }
+}
+
+// --- Edit-apply planning (pure, unit-tested) ---------------------------------
+
+/**
+ * Decision for applying one queued edit to a file's current content. Pulled out
+ * of `applyOne` so the splice — the part that actually corrupted files — is pure
+ * and testable without a DB or filesystem. Mirrors how opencode/cline/qwen keep
+ * their matchers fail-closed and idempotent.
+ */
+export type EditPlan =
+  | { kind: 'apply'; updated: string }
+  | { kind: 'noop'; reason: 'identical' | 'already-applied' }
+  | { kind: 'ambiguous'; count: number }
+  | { kind: 'not_found' };
+
+/**
+ * Decide how (or whether) to apply an `old → new` edit to `content`.
+ *
+ * Idempotency is the whole point here: a queued edit can legitimately be
+ * re-applied (a local model re-emits the same tool call; a turn is retried; the
+ * same change sits in the queue twice). A naive splice stamps the new text again
+ * each time — the 2–3× block duplication. Two guards make re-application a no-op:
+ *
+ *  - already-applied (anchored insert): when `new` is `old` + an appended block
+ *    (`old="anchor"`, `new="anchor\n<block>"`), `old` still matches uniquely after
+ *    the first apply, so a second apply would duplicate `<block>`. If the full
+ *    `new` text is already present at the match site, the edit is already applied.
+ *  - already-applied (old gone): if `old` can't be located but `new` is already
+ *    in the file, the change landed on a prior pass — treat as a no-op, not an error.
+ *  - identical: the splice would not change the file.
+ *
+ * Anything ambiguous or genuinely absent fails CLOSED so the caller surfaces a
+ * correctable error instead of writing a guess.
+ */
+export function planEdit(content: string, oldStr: string, newStr: string): EditPlan {
+  const match = locateMatch(content, oldStr);
+
+  if (match.kind === 'ambiguous') return { kind: 'ambiguous', count: match.count };
+
+  if (match.kind === 'not_found') {
+    if (newStr.length > 0 && content.includes(newStr)) {
+      return { kind: 'noop', reason: 'already-applied' };
+    }
+    return { kind: 'not_found' };
+  }
+
+  const updated = content.slice(0, match.start) + newStr + content.slice(match.end);
+  // No-change splice first (covers old === new), then the anchored re-stamp guard:
+  // the full replacement already sits at the match site (re-emitted anchored insert).
+  if (updated === content) return { kind: 'noop', reason: 'identical' };
+  if (content.slice(match.start, match.start + newStr.length) === newStr) {
+    return { kind: 'noop', reason: 'already-applied' };
+  }
+  return { kind: 'apply', updated };
+}
+
 // --- Types -------------------------------------------------------------------
 
 export interface PendingChange {
@@ -47,6 +158,13 @@ export async function queueEdit(
   const resolved = resolveWritePath(projectRoot, filePath);
   const diff = JSON.stringify({ old: oldString, new: newString });
 
+  // Idempotent queue: collapse an identical edit that is still pending. Local
+  // quantized models re-emit the same edit_file call within a turn, and a retried
+  // turn re-queues — each duplicate row would apply and stamp another copy. One
+  // pending row per (session, file, operation, diff) is enough.
+  const existing = await findPendingDuplicate(sql, sessionId, resolved, 'edit', diff);
+  if (existing) return existing;
+
   const [row] = await sql<PendingChange[]>`
     INSERT INTO pending_changes (session_id, task_id, file_path, operation, diff, agent)
     VALUES (${sessionId}, ${taskId}, ${resolved}, 'edit', ${diff}, ${agent})
@@ -55,6 +173,28 @@ export async function queueEdit(
   return row!;
 }
 
+/** Return an identical still-pending change for this (session, file, op, diff),
+ *  or undefined. Used to keep the queue idempotent against re-emitted edits. */
+async function findPendingDuplicate(
+  sql: Sql,
+  sessionId: string,
+  resolvedPath: string,
+  operation: 'create' | 'edit' | 'delete',
+  diff: string,
+): Promise<PendingChange | undefined> {
+  const [row] = await sql<PendingChange[]>`
+    SELECT * FROM pending_changes
+    WHERE session_id = ${sessionId}
+      AND file_path = ${resolvedPath}
+      AND operation = ${operation}
+      AND diff = ${diff}
+      AND status = 'pending'
+    ORDER BY created_at ASC
+    LIMIT 1
+  `;
+  return row;
+}
+
 export async function queueCreate(
   sql: Sql,
   sessionId: string,
@@ -68,6 +208,9 @@ export async function queueCreate(
 ): Promise<PendingChange> {
   const resolved = resolveWritePath(projectRoot, filePath);
 
+  const existing = await findPendingDuplicate(sql, sessionId, resolved, 'create', content);
+  if (existing) return existing;
+
   const [row] = await sql<PendingChange[]>`
     INSERT INTO pending_changes (session_id, task_id, file_path, operation, diff, agent)
     VALUES (${sessionId}, ${taskId}, ${resolved}, 'create', ${content}, ${agent})
@@ -87,6 +230,9 @@ export async function queueDelete(
 ): Promise<PendingChange> {
   const resolved = resolveWritePath(projectRoot, filePath);
 
+  const existing = await findPendingDuplicate(sql, sessionId, resolved, 'delete', '');
+  if (existing) return existing;
+
   const [row] = await sql<PendingChange[]>`
     INSERT INTO pending_changes (session_id, task_id, file_path, operation, diff, agent)
     VALUES (${sessionId}, ${taskId}, ${resolved}, 'delete', '', ${agent})
@@ -110,48 +256,60 @@ export async function applyOne(
   }
 
   try {
-    // Re-validate path in case projectRoot has shifted
+    return await withFileLock(change.file_path, async () => {
-    resolveWritePath(projectRoot, change.file_path);
+      // Re-validate path in case projectRoot has shifted
+      resolveWritePath(projectRoot, change.file_path);
 
-    switch (change.operation) {
+      switch (change.operation) {
-      case 'create': {
+        case 'create': {
-        await mkdir(dirname(change.file_path), { recursive: true });
+          await mkdir(dirname(change.file_path), { recursive: true });
-        await writeFile(change.file_path, change.diff, 'utf8');
+          await writeFileAtomic(change.file_path, change.diff);
-        break;
+          break;
-      }
-      case 'edit': {
-        const { old: oldStr, new: newStr } = JSON.parse(change.diff) as { old: string; new: string };
-        const content = await readFile(change.file_path, 'utf8');
-        const match = locateMatch(content, oldStr);
-        if (match.kind === 'ambiguous') {
-          throw new Error(
-            `old_string matches ${match.count} locations — add surrounding context to disambiguate`,
-          );
         }
-        if (match.kind === 'not_found') {
+        case 'edit': {
-          throw new Error(
+          const { old: oldStr, new: newStr } = JSON.parse(change.diff) as { old: string; new: string };
-            'old_string not found in file (even fuzzily) — file may have changed since the edit was queued',
+          const raw = await readFile(change.file_path, 'utf8');
-          );
+          // Normalize to LF for matching, then write back in the file's native EOL
+          // so an LF-emitting model doesn't leave a CRLF file with mixed endings.
+          const eol = detectEol(raw);
+          const toLf = (t: string) => t.replaceAll('\r\n', '\n');
+          const plan = planEdit(toLf(raw), toLf(oldStr), toLf(newStr));
+          if (plan.kind === 'ambiguous') {
+            throw new Error(
+              `old_string matches ${plan.count} locations — add surrounding context to disambiguate`,
+            );
+          }
+          if (plan.kind === 'not_found') {
+            throw new Error(
+              'old_string not found in file (even fuzzily) — file may have changed since the edit was queued',
+            );
+          }
+          if (plan.kind === 'apply') {
+            const out = eol === '\r\n' ? plan.updated.replaceAll('\n', '\r\n') : plan.updated;
+            await writeFileAtomic(change.file_path, out);
+          } else {
+            // noop: the edit is already applied (re-emitted / retried) or a no-change.
+            // Mark it applied without rewriting so it can't stamp a duplicate.
+            console.log(`[pending] edit ${change.file_path} is a no-op (${plan.reason}) — not rewriting`);
+          }
+          break;
         }
-        const updated = content.slice(0, match.start) + newStr + content.slice(match.end);
+        case 'delete': {
-        await writeFile(change.file_path, updated, 'utf8');
+          // Stash current content in diff for potential rewind
-        break;
+          try {
-      }
+            const existing = await readFile(change.file_path, 'utf8');
-      case 'delete': {
+            await sql`UPDATE pending_changes SET diff = ${existing} WHERE id = ${changeId}`;
-        // Stash current content in diff for potential rewind
+          } catch {
-        try {
+            // File may already be gone — proceed with status update
-          const existing = await readFile(change.file_path, 'utf8');
+          }
-          await sql`UPDATE pending_changes SET diff = ${existing} WHERE id = ${changeId}`;
+          await unlink(change.file_path);
-        } catch {
+          break;
-          // File may already be gone — proceed with status update
         }
-        await unlink(change.file_path);
-        break;
       }
-    }
 
-    await sql`UPDATE pending_changes SET status = 'applied' WHERE id = ${changeId}`;
+      await sql`UPDATE pending_changes SET status = 'applied' WHERE id = ${changeId}`;
-    return { id: change.id, file_path: change.file_path, operation: change.operation, success: true };
+      return { id: change.id, file_path: change.file_path, operation: change.operation, success: true };
+    });
   } catch (err) {
     const message = err instanceof Error ? err.message : String(err);
     return { id: change.id, file_path: change.file_path, operation: change.operation, success: false, error: message };
@@ -220,13 +378,13 @@ export async function rewindOne(
           );
         }
         const reverted = content.slice(0, match.start) + oldStr + content.slice(match.end);
-        await writeFile(change.file_path, reverted, 'utf8');
+        await writeFileAtomic(change.file_path, reverted);
         break;
       }
       case 'delete': {
         // Reverse a delete: recreate the file (diff holds the original content stashed at apply time)
         await mkdir(dirname(change.file_path), { recursive: true });
-        await writeFile(change.file_path, change.diff, 'utf8');
+        await writeFileAtomic(change.file_path, change.diff);
         break;
       }
     }

apps/coder/src/services/provider-manifest.ts

@@ -32,6 +32,18 @@ const QWEN_PTY_MODES: ProviderMode[] = [
   { id: 'yolo', label: 'YOLO', description: 'Auto-approve all tools', isUnattended: true },
 ];
 
+// Native BooCode (llama-swap) has no agent-native mode vocabulary, so we define
+// one that matches the unified permission ladder. `bypass` is the only mode that
+// changes behavior (auto-apply staged edits after the turn — dispatcher.ts);
+// `plan` falls back to `ask` semantics for native (writes still stage to the
+// pending-changes queue). External agents map the same three unified modes onto
+// THEIR native ids via the `plan`-id / default / `isUnattended` shape.
+const BOOCODE_MODES: ProviderMode[] = [
+  { id: 'plan', label: 'Plan', description: 'Read-only analysis (native BooCode falls back to Ask)' },
+  { id: 'ask', label: 'Ask Permission', description: 'Stage edits to the pending-changes queue for review' },
+  { id: 'bypass', label: 'Bypass', description: 'Auto-apply edits to disk after the turn', isUnattended: true },
+];
+
 const CLAUDE_THINKING = [
   { id: 'low', label: 'Low' },
   { id: 'medium', label: 'Medium' },
@@ -41,6 +53,10 @@ const CLAUDE_THINKING = [
 ];
 
 export const PROVIDER_MANIFEST: Record<string, ProviderManifestEntry> = {
+  boocode: {
+    defaultModeId: 'ask',
+    modes: BOOCODE_MODES,
+  },
   claude: {
     defaultModeId: 'default',
     modes: CLAUDE_MODES,

apps/coder/src/services/provider-snapshot.ts

@@ -122,12 +122,14 @@ async function buildProviderEntry(
     };
   }
 
-  // 2. Native boocode → always ready (llama-swap models).
+  // 2. Native boocode → always ready (llama-swap models). Exposes the unified
+  // permission modes (plan/ask/bypass) so the composer's permission picker works
+  // for native BooCode too; `bypass` auto-applies staged edits (dispatcher.ts).
   if (isNative) {
     return {
       name, label: resolved.label, transport, status: 'ready',
-      enabled: true, installed: true, models: withConfigModels(llamaModels), modes: [],
+      enabled: true, installed: true, models: withConfigModels(llamaModels),
-      defaultModeId: null, commands: manifestCommands,
+      modes: fallbackModes, defaultModeId, commands: manifestCommands,
     };
   }
 

apps/coder/src/services/token-analysis/__tests__/analyzer.test.ts

@@ -0,0 +1,29 @@
+import { describe, it, expect } from 'vitest';
+import { analyzeMessages } from '../analyzer.js';
+
+describe('analyzeMessages', () => {
+  it('classifies user messages', () => {
+    const breakdown = analyzeMessages([{ role: 'user', content: 'hello world' }]);
+    expect(breakdown.user).toBeGreaterThan(0);
+    expect(breakdown.total).toBe(breakdown.user);
+  });
+
+  it('counts tool calls', () => {
+    const parts = [
+      { role: 'assistant', content: 'using grep', tool_calls: [{ id: '1', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: '{"files":[]}', tool_call_id: '1' },
+    ];
+    const breakdown = analyzeMessages(parts);
+    expect(breakdown.tools).toBeGreaterThan(0);
+    expect(breakdown.assistant).toBeGreaterThan(0);
+  });
+
+  it('separates reasoning tokens', () => {
+    const parts = [
+      { role: 'assistant', content: 'short answer', reasoning_parts: [{ text: 'long chain of thought reasoning here' }] },
+    ];
+    const breakdown = analyzeMessages(parts);
+    expect(breakdown.reasoning).toBeGreaterThan(0);
+    expect(breakdown.assistant).toBeLessThan(breakdown.reasoning);
+  });
+});

apps/coder/src/services/token-analysis/__tests__/persist.test.ts

@@ -0,0 +1,10 @@
+import { describe, it, expect } from 'vitest';
+
+describe('persistTaskBreakdown', () => {
+  it('exports functions', async () => {
+    const mod = await import('../persist.js');
+    expect(typeof mod.persistTaskBreakdown).toBe('function');
+    expect(typeof mod.getTaskBreakdown).toBe('function');
+    expect(typeof mod.analyzeAndPersistTaskBreakdown).toBe('function');
+  });
+});

apps/coder/src/services/token-analysis/analyzer.ts

@@ -0,0 +1,60 @@
+// TokenScope analyzer — classifies message parts into category breakdown.
+// Ported from opencode-tokenscope (MIT).
+
+export interface TokenBreakdown {
+  system: number;
+  user: number;
+  assistant: number;
+  tools: number;
+  reasoning: number;
+  total: number;
+}
+
+const CHARS_PER_TOKEN = 4;
+
+function estimateTokens(text: string): number {
+  return Math.ceil(text.length / CHARS_PER_TOKEN);
+}
+
+export function analyzeMessages(parts: any[]): TokenBreakdown {
+  const breakdown: TokenBreakdown = { system: 0, user: 0, assistant: 0, tools: 0, reasoning: 0, total: 0 };
+
+  for (const part of parts) {
+    const role = part.role ?? '';
+    const content = part.content ?? '';
+    const tokens = estimateTokens(content);
+
+    switch (role) {
+      case 'system':
+        breakdown.system += tokens;
+        break;
+      case 'user':
+        breakdown.user += tokens;
+        break;
+      case 'assistant':
+        breakdown.assistant += tokens;
+        if (part.tool_calls) {
+          for (const tc of part.tool_calls) {
+            breakdown.tools += estimateTokens(JSON.stringify(tc));
+          }
+        }
+        break;
+      case 'tool':
+        breakdown.tools += tokens;
+        break;
+      default:
+        breakdown.assistant += tokens;
+    }
+
+    if (part.reasoning_parts) {
+      for (const rp of part.reasoning_parts) {
+        const rTokens = estimateTokens(rp.text ?? '');
+        breakdown.reasoning += rTokens;
+        breakdown.assistant -= rTokens;
+      }
+    }
+  }
+
+  breakdown.total = breakdown.system + breakdown.user + breakdown.assistant + breakdown.tools + breakdown.reasoning;
+  return breakdown;
+}

apps/coder/src/services/token-analysis/persist.ts

@@ -0,0 +1,35 @@
+// TokenScope persistence — writes breakdown to task records.
+import type { Sql } from '../../db.js';
+import type { TokenBreakdown } from './analyzer.js';
+
+export async function persistTaskBreakdown(
+  sql: Sql,
+  taskId: string,
+  breakdown: TokenBreakdown,
+): Promise<void> {
+  await sql`
+    UPDATE tasks SET token_breakdown = ${sql.json(breakdown as never)}
+    WHERE id = ${taskId}
+  `;
+}
+
+export async function getTaskBreakdown(
+  sql: Sql,
+  taskId: string,
+): Promise<TokenBreakdown | null> {
+  const rows = await sql<{ token_breakdown: any }[]>`
+    SELECT token_breakdown FROM tasks WHERE id = ${taskId}
+  `;
+  return rows[0]?.token_breakdown ?? null;
+}
+
+export async function analyzeAndPersistTaskBreakdown(
+  sql: Sql,
+  taskId: string,
+  parts: any[],
+): Promise<TokenBreakdown> {
+  const { analyzeMessages } = await import('./analyzer.js');
+  const breakdown = analyzeMessages(parts);
+  await persistTaskBreakdown(sql, taskId, breakdown);
+  return breakdown;
+}

apps/coder/src/services/tools/__tests__/inference_context.test.ts

@@ -0,0 +1,38 @@
+import { describe, it, expect } from 'vitest';
+import { runWithInferenceContext, getInferenceContext } from '../inference_context.js';
+import type { Sql } from '../../../db.js';
+
+const fakeSql = {} as unknown as Sql;
+
+describe('inference context (AsyncLocalStorage isolation)', () => {
+  it('throws when read outside a run', () => {
+    expect(() => getInferenceContext()).toThrow(/outside inference context/);
+  });
+
+  it('keeps each run its own context across overlapping awaits', async () => {
+    // The race the global `let current` had: run B starts (and would overwrite a
+    // shared global) while run A is awaiting. After A resumes it must still read
+    // its OWN sessionId, not B's.
+    const run = (id: string, delay: number) =>
+      runWithInferenceContext({ sql: fakeSql, sessionId: id, taskId: null }, async () => {
+        await new Promise((r) => setTimeout(r, delay));
+        return getInferenceContext().sessionId;
+      });
+
+    const [a, b] = await Promise.all([run('A', 20), run('B', 5)]);
+    expect(a).toBe('A');
+    expect(b).toBe('B');
+  });
+
+  it('carries permissionMode and taskId per run', async () => {
+    const result = await runWithInferenceContext(
+      { sql: fakeSql, sessionId: 's1', taskId: 't1', permissionMode: 'bypass' },
+      async () => {
+        await Promise.resolve();
+        const ctx = getInferenceContext();
+        return { taskId: ctx.taskId, mode: ctx.permissionMode };
+      },
+    );
+    expect(result).toEqual({ taskId: 't1', mode: 'bypass' });
+  });
+});

apps/coder/src/services/tools/apply_pending.ts

@@ -26,6 +26,15 @@ export const applyPendingTool: ToolDef<ApplyPendingInputT> = {
     },
   },
   async execute(_input: ApplyPendingInputT, projectRoot: string, context: ToolContext): Promise<unknown> {
+    // Under Ask (and Plan) the human approves via the Pending Changes panel — the
+    // agent must not auto-apply. Bypass and legacy (undefined) may apply.
+    if (context.permissionMode === 'ask' || context.permissionMode === 'plan') {
+      return {
+        status: 'denied',
+        message:
+          'Permission mode is Ask — staged changes must be approved by the user in the Pending Changes panel, not applied by the agent.',
+      };
+    }
     const results = await applyAll(context.sql, context.sessionId, projectRoot);
     const succeeded = results.filter((r) => r.success).length;
     const failed = results.filter((r) => !r.success).length;

apps/coder/src/services/tools/create_file.ts

@@ -1,6 +1,7 @@
 import { z } from 'zod';
 import type { ToolDef, ToolContext } from './types.js';
 import { queueCreate } from '../pending_changes.js';
+import { denyReadOnly, finalizeWrite } from './write-gate.js';
 
 const CreateFileInput = z.object({
   file_path: z.string().min(1),
@@ -32,6 +33,7 @@ export const createFileTool: ToolDef<CreateFileInputT> = {
     },
   },
   async execute(input: CreateFileInputT, projectRoot: string, context: ToolContext): Promise<unknown> {
+    if (context.permissionMode === 'plan') return denyReadOnly('create_file');
     const change = await queueCreate(
       context.sql,
       context.sessionId,
@@ -40,12 +42,11 @@ export const createFileTool: ToolDef<CreateFileInputT> = {
       input.content,
       projectRoot,
     );
-    return {
+    return finalizeWrite(
-      status: 'queued',
+      context,
-      change_id: change.id,
+      projectRoot,
-      file_path: change.file_path,
+      change,
-      operation: 'create',
+      `File creation queued: ${change.file_path}. Use apply_pending to write changes to disk.`,
-      message: `File creation queued: ${change.file_path}. Use apply_pending to write changes to disk.`,
+    );
-    };
   },
 };

apps/coder/src/services/tools/delete_file.ts

@@ -1,6 +1,7 @@
 import { z } from 'zod';
 import type { ToolDef, ToolContext } from './types.js';
 import { queueDelete } from '../pending_changes.js';
+import { denyReadOnly, finalizeWrite } from './write-gate.js';
 
 const DeleteFileInput = z.object({
   file_path: z.string().min(1),
@@ -30,6 +31,7 @@ export const deleteFileTool: ToolDef<DeleteFileInputT> = {
     },
   },
   async execute(input: DeleteFileInputT, projectRoot: string, context: ToolContext): Promise<unknown> {
+    if (context.permissionMode === 'plan') return denyReadOnly('delete_file');
     const change = await queueDelete(
       context.sql,
       context.sessionId,
@@ -37,12 +39,11 @@ export const deleteFileTool: ToolDef<DeleteFileInputT> = {
       input.file_path,
       projectRoot,
     );
-    return {
+    return finalizeWrite(
-      status: 'queued',
+      context,
-      change_id: change.id,
+      projectRoot,
-      file_path: change.file_path,
+      change,
-      operation: 'delete',
+      `File deletion queued: ${change.file_path}. Use apply_pending to write changes to disk.`,
-      message: `File deletion queued: ${change.file_path}. Use apply_pending to write changes to disk.`,
+    );
-    };
   },
 };

apps/coder/src/services/tools/edit_file.ts

@@ -1,6 +1,7 @@
 import { z } from 'zod';
 import type { ToolDef, ToolContext } from './types.js';
 import { queueEdit } from '../pending_changes.js';
+import { denyReadOnly, finalizeWrite } from './write-gate.js';
 
 const EditFileInput = z.object({
   file_path: z.string().min(1),
@@ -34,6 +35,7 @@ export const editFileTool: ToolDef<EditFileInputT> = {
     },
   },
   async execute(input: EditFileInputT, projectRoot: string, context: ToolContext): Promise<unknown> {
+    if (context.permissionMode === 'plan') return denyReadOnly('edit_file');
     const change = await queueEdit(
       context.sql,
       context.sessionId,
@@ -43,12 +45,11 @@ export const editFileTool: ToolDef<EditFileInputT> = {
       input.new_string,
       projectRoot,
     );
-    return {
+    return finalizeWrite(
-      status: 'queued',
+      context,
-      change_id: change.id,
+      projectRoot,
-      file_path: change.file_path,
+      change,
-      operation: 'edit',
+      `Edit queued for ${change.file_path}. Use apply_pending to write changes to disk.`,
-      message: `Edit queued for ${change.file_path}. Use apply_pending to write changes to disk.`,
+    );
-    };
   },
 };

apps/coder/src/services/tools/index.ts

@@ -7,6 +7,9 @@ import { rewindTool } from './rewind.js';
 import { newTaskTool } from './new_task.js';
 import { listTasksTool } from './list_tasks.js';
 import { checkTaskStatusTool } from './check_task_status.js';
+import { lspDiagnosticsTool } from './lsp_diagnostics.js';
+import { lspGotoDefinitionTool } from './lsp_goto_definition.js';
+import { lspFindReferencesTool } from './lsp_find_references.js';
 
 export type { ToolDef, ToolContext, ToolJsonSchema } from './types.js';
 
@@ -26,4 +29,16 @@ export const WRITE_TOOLS: readonly ToolDef<any>[] = [
   checkTaskStatusTool,
 ];
 
-export { editFileTool, createFileTool, deleteFileTool, applyPendingTool, rewindTool, newTaskTool, listTasksTool, checkTaskStatusTool };
+// Read-only agent tools for code intelligence.
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+export const READ_TOOLS: readonly ToolDef<any>[] = [
+  lspDiagnosticsTool,
+  lspGotoDefinitionTool,
+  lspFindReferencesTool,
+];
+
+export {
+  editFileTool, createFileTool, deleteFileTool, applyPendingTool, rewindTool,
+  newTaskTool, listTasksTool, checkTaskStatusTool,
+  lspDiagnosticsTool, lspGotoDefinitionTool, lspFindReferencesTool,
+};

apps/coder/src/services/tools/inference_context.ts

@@ -1,36 +1,49 @@
+import { AsyncLocalStorage } from 'node:async_hooks';
 import type { Sql } from '../../db.js';
+import type { PermissionMode } from './types.js';
 
 /**
- * Module-level inference context for write tools.
+ * Per-run inference context for write tools.
  *
- * Set via `setInferenceContext()` before each inference run starts.
+ * Write tools need ambient state (sql, sessionId, the permission gate) that the
- * Write tools read it via `getInferenceContext()` during execute.
+ * BooChat tool-phase `execute(input, projectRoot, extraRoots?)` signature can't
- * Same pattern as BooChat's `loadConfig()` singleton — tools need
+ * carry. This used to be a single module-level `let current` — but the inference
- * ambient state that can't be threaded through the tool-phase execute
+ * runner's `enqueue()` is fire-and-forget, so two overlapping runs (a user
- * signature (which is `execute(input, projectRoot, extraRoots?)`).
+ * message racing a dispatcher-polled native task; two chat tabs streaming) would
+ * clobber each other's context, and `cancel()` cleared it for ALL in-flight runs.
+ *
+ * AsyncLocalStorage gives each run its own context: `enqueue()` starts its async
+ * loop synchronously inside `runWithInferenceContext`, so the store propagates
+ * through every awaited tool execution in that run — and only that run.
  */
 
 export interface InferenceContext {
   sql: Sql;
   sessionId: string;
   taskId: string | null;
+  /** Native-BooCode permission gate, set per run from the request/task mode. */
+  permissionMode?: PermissionMode;
 }
 
-let current: InferenceContext | null = null;
+const storage = new AsyncLocalStorage<InferenceContext>();
 
-export function setInferenceContext(ctx: InferenceContext): void {
+/**
-  current = ctx;
+ * Bind `ctx` for the duration of the (possibly detached) async chain `fn` starts.
-}
+ * The inference runner kicks off its loop synchronously within this call, so all
-
+ * downstream `await`s — including write-tool `execute` via the adapter — read the
-export function clearInferenceContext(): void {
+ * same store. Concurrent runs each get their own; nothing is shared or cleared
-  current = null;
+ * out from under an in-flight run.
+ */
+export function runWithInferenceContext<T>(ctx: InferenceContext, fn: () => T): T {
+  return storage.run(ctx, fn);
 }
 
 export function getInferenceContext(): InferenceContext {
-  if (!current) {
+  const ctx = storage.getStore();
+  if (!ctx) {
     throw new Error(
-      'Write tool called outside inference context — setInferenceContext() was not called before this run',
+      'Write tool called outside inference context — runWithInferenceContext() did not wrap this run',
     );
   }
-  return current;
+  return ctx;
 }

apps/coder/src/services/tools/lsp_diagnostics.ts

@@ -0,0 +1,48 @@
+import { z } from 'zod';
+import { readFile } from 'node:fs/promises';
+import type { ToolDef, ToolContext } from './types.js';
+import { resolveWritePath } from '../write_guard.js';
+import { lspManager } from '../lsp/server-manager.js';
+import { getDiagnostics } from '../lsp/operations.js';
+
+const LspDiagnosticsInput = z.object({
+  file_path: z.string().describe('Path to the file to check for diagnostics'),
+});
+
+type InputT = z.infer<typeof LspDiagnosticsInput>;
+
+export const lspDiagnosticsTool: ToolDef<InputT> = {
+  name: 'lsp_diagnostics',
+  description: 'Get TypeScript/JavaScript diagnostics (errors, warnings) for a file. Returns diagnostic messages with severity and location.',
+  inputSchema: LspDiagnosticsInput,
+  jsonSchema: {
+    type: 'function',
+    function: {
+      name: 'lsp_diagnostics',
+      description: 'Get TypeScript/JavaScript diagnostics for a file',
+      parameters: {
+        type: 'object',
+        properties: {
+          file_path: { type: 'string', description: 'Path to the file' },
+        },
+        required: ['file_path'],
+      },
+    },
+  },
+
+  async execute(input: InputT, projectRoot: string, _context: ToolContext): Promise<unknown> {
+    const resolved = await resolveWritePath(projectRoot, input.file_path);
+    const content = await readFile(resolved, 'utf8');
+    const client = await lspManager.getClient(resolved);
+    if (!client) return { error: 'Unsupported file type for LSP diagnostics' };
+
+    const diagnostics = await getDiagnostics(client, resolved, content);
+    if (diagnostics.length === 0) return { result: 'No diagnostics found.' };
+
+    const lines = diagnostics.map((d) => {
+      const sev = ['', 'error', 'warning', 'info', 'hint'][d.severity] ?? 'unknown';
+      return `[${sev}] line ${d.range.start.line + 1}:${d.range.start.character + 1} - ${d.message}`;
+    });
+    return { result: lines.join('\n') };
+  },
+};

apps/coder/src/services/tools/lsp_find_references.ts

@@ -0,0 +1,49 @@
+import { z } from 'zod';
+import { readFile } from 'node:fs/promises';
+import type { ToolDef, ToolContext } from './types.js';
+import { resolveWritePath } from '../write_guard.js';
+import { lspManager } from '../lsp/server-manager.js';
+import { findReferences } from '../lsp/operations.js';
+
+const LspFindReferencesInput = z.object({
+  file_path: z.string().describe('Path to the source file'),
+  line: z.number().int().nonnegative().describe('0-based line number'),
+  character: z.number().int().nonnegative().describe('0-based character offset'),
+});
+
+type InputT = z.infer<typeof LspFindReferencesInput>;
+
+export const lspFindReferencesTool: ToolDef<InputT> = {
+  name: 'lsp_find_references',
+  description: 'Find all references to a symbol at a given position in a file.',
+  inputSchema: LspFindReferencesInput,
+  jsonSchema: {
+    type: 'function',
+    function: {
+      name: 'lsp_find_references',
+      description: 'Find all references to symbol at position',
+      parameters: {
+        type: 'object',
+        properties: {
+          file_path: { type: 'string' },
+          line: { type: 'number' },
+          character: { type: 'number' },
+        },
+        required: ['file_path', 'line', 'character'],
+      },
+    },
+  },
+
+  async execute(input: InputT, projectRoot: string, _context: ToolContext): Promise<unknown> {
+    const resolved = await resolveWritePath(projectRoot, input.file_path);
+    const content = await readFile(resolved, 'utf8');
+    const client = await lspManager.getClient(resolved);
+    if (!client) return { error: 'Unsupported file type' };
+
+    const refs = await findReferences(client, resolved, content, input.line, input.character);
+    if (refs.length === 0) return { result: 'No references found.' };
+
+    const lines = refs.map((r) => `${r.uri}:${r.range.start.line + 1}:${r.range.start.character + 1}`);
+    return { result: `Found ${refs.length} reference(s):\n${lines.join('\n')}` };
+  },
+};

apps/coder/src/services/tools/lsp_goto_definition.ts

@@ -0,0 +1,48 @@
+import { z } from 'zod';
+import { readFile } from 'node:fs/promises';
+import type { ToolDef, ToolContext } from './types.js';
+import { resolveWritePath } from '../write_guard.js';
+import { lspManager } from '../lsp/server-manager.js';
+import { gotoDefinition } from '../lsp/operations.js';
+
+const LspGotoDefinitionInput = z.object({
+  file_path: z.string().describe('Path to the source file'),
+  line: z.number().int().nonnegative().describe('0-based line number'),
+  character: z.number().int().nonnegative().describe('0-based character offset'),
+});
+
+type InputT = z.infer<typeof LspGotoDefinitionInput>;
+
+export const lspGotoDefinitionTool: ToolDef<InputT> = {
+  name: 'lsp_goto_definition',
+  description: 'Find the definition of a symbol at a given position in a file.',
+  inputSchema: LspGotoDefinitionInput,
+  jsonSchema: {
+    type: 'function',
+    function: {
+      name: 'lsp_goto_definition',
+      description: 'Find definition of symbol at position',
+      parameters: {
+        type: 'object',
+        properties: {
+          file_path: { type: 'string' },
+          line: { type: 'number' },
+          character: { type: 'number' },
+        },
+        required: ['file_path', 'line', 'character'],
+      },
+    },
+  },
+
+  async execute(input: InputT, projectRoot: string, _context: ToolContext): Promise<unknown> {
+    const resolved = await resolveWritePath(projectRoot, input.file_path);
+    const content = await readFile(resolved, 'utf8');
+    const client = await lspManager.getClient(resolved);
+    if (!client) return { error: 'Unsupported file type' };
+
+    const loc = await gotoDefinition(client, resolved, content, input.line, input.character);
+    if (!loc) return { result: 'No definition found.' };
+
+    return { result: `Defined at ${loc.uri}:${loc.range.start.line + 1}:${loc.range.start.character + 1}` };
+  },
+};

apps/coder/src/services/tools/new_task.ts

@@ -6,6 +6,7 @@ const NewTaskInput = z.object({
   input: z.string().min(1).describe('Task description for the child subtask'),
   agent: z.string().optional().describe('Optional: dispatch to a specific agent'),
   model: z.string().optional().describe('Optional: model override for the subtask'),
+  background: z.boolean().optional().describe('If true, return immediately without blocking on completion'),
 });
 
 type NewTaskInputT = z.infer<typeof NewTaskInput>;
@@ -30,6 +31,7 @@ export const newTaskTool: ToolDef<NewTaskInputT> = {
           input: { type: 'string', description: 'Task description for the child subtask' },
           agent: { type: 'string', description: 'Optional: dispatch to a specific agent' },
           model: { type: 'string', description: 'Optional: model override for the subtask' },
+          background: { type: 'boolean', description: 'If true, returns immediately without waiting' },
         },
         required: ['input'],
       },
@@ -50,6 +52,7 @@ export const newTaskTool: ToolDef<NewTaskInputT> = {
       return { error: 'Cannot determine project_id from current session' };
     }
 
+    const isBg = input.background === true;
     const [task] = await sql<{ id: string; state: string }[]>`
       INSERT INTO tasks (project_id, parent_task_id, input, agent, model)
       VALUES (${session.project_id}, ${currentTaskId}, ${input.input}, ${input.agent ?? null}, ${input.model ?? null})
@@ -57,9 +60,12 @@ export const newTaskTool: ToolDef<NewTaskInputT> = {
     `;
 
     return {
-      message: `Subtask created (id: ${task!.id}). It will run in isolation. Use check_task_status to monitor.`,
+      message: isBg
+        ? `Background subtask created (id: ${task!.id}). It will continue independently.`
+        : `Subtask created (id: ${task!.id}). It will run in isolation. Use check_task_status to monitor.`,
       task_id: task!.id,
       state: task!.state,
+      background: isBg,
     };
   },
 };

apps/coder/src/services/tools/types.ts

@@ -1,6 +1,22 @@
 import type { z } from 'zod';
 import type { Sql } from '../../db.js';
 
+/**
+ * Unified permission ladder for native BooCode inference. Gates the write tools:
+ *   plan   — read-only: create/edit/delete are denied (no staging).
+ *   ask    — stage to the pending-changes queue; `apply_pending` is denied so the
+ *            agent cannot self-apply (the human approves via the Diff panel).
+ *   bypass — apply each write immediately (no queue, no approval).
+ * Undefined preserves the historical behavior (stage + `apply_pending` allowed).
+ */
+export type PermissionMode = 'plan' | 'ask' | 'bypass';
+
+/** Narrow a raw task/request mode id to a unified PermissionMode, else undefined
+ *  (e.g. an external agent's native mode id, or null). */
+export function asPermissionMode(id: string | null | undefined): PermissionMode | undefined {
+  return id === 'plan' || id === 'ask' || id === 'bypass' ? id : undefined;
+}
+
 export interface ToolJsonSchema {
   type: 'function';
   function: {
@@ -21,6 +37,8 @@ export interface ToolContext {
   sql: Sql;
   sessionId: string;
   taskId: string | null;
+  /** Native-BooCode permission gate for write tools (undefined = legacy behavior). */
+  permissionMode?: PermissionMode;
 }
 
 export interface ToolDef<TInput> {

apps/coder/src/services/tools/write-gate.ts

@@ -0,0 +1,53 @@
+/**
+ * Permission-gate helpers for native BooCode write tools. The gate comes from
+ * the per-run inference context (`ToolContext.permissionMode`):
+ *   plan   — deny the write (read-only); nothing is staged.
+ *   bypass — apply the staged change immediately (no queue, no approval).
+ *   ask / undefined — leave it in the pending-changes queue for review.
+ */
+import type { ToolContext } from './types.js';
+import { applyOne } from '../pending_changes.js';
+
+/** Result returned when a write is denied under Plan (read-only) mode. */
+export function denyReadOnly(operation: string): unknown {
+  return {
+    status: 'denied',
+    operation,
+    message: `Read-only (Plan) permission mode — ${operation} is not permitted. Switch to Ask or Bypass to make changes.`,
+  };
+}
+
+/** Finalize a just-staged change per the permission gate: apply now under Bypass,
+ *  otherwise return it as queued for the human to approve. */
+export async function finalizeWrite(
+  context: ToolContext,
+  projectRoot: string,
+  change: { id: string; file_path: string; operation: string },
+  queuedHint: string,
+): Promise<unknown> {
+  if (context.permissionMode === 'bypass') {
+    const res = await applyOne(context.sql, change.id, projectRoot);
+    console.log(
+      `[write-gate] bypass apply ${change.operation} ${change.file_path} -> ${res.success ? 'applied' : 'FAILED: ' + (res.error ?? '?')}`,
+    );
+    return {
+      status: res.success ? 'applied' : 'failed',
+      change_id: change.id,
+      file_path: change.file_path,
+      operation: change.operation,
+      message: res.success
+        ? `${change.operation} applied to ${change.file_path}.`
+        : `Apply failed for ${change.file_path}: ${res.error ?? 'unknown error'}. Left in the pending queue.`,
+    };
+  }
+  console.log(
+    `[write-gate] ${context.permissionMode ?? 'legacy'} queued ${change.operation} ${change.file_path}`,
+  );
+  return {
+    status: 'queued',
+    change_id: change.id,
+    file_path: change.file_path,
+    operation: change.operation,
+    message: queuedHint,
+  };
+}

apps/server/CLAUDE.md

@@ -17,6 +17,8 @@
   - **Tools have NO `execute` field.** BooCode dispatches tools in tool-phase.ts, not the AI SDK loop — only `description` + `inputSchema: jsonSchema(parameters)`.
   - **`includeUsage: true` MUST be set on `createOpenAICompatible`** in `provider.ts`. The adapter defaults it false → no `stream_options.include_usage` → llama-swap emits no usage block → `result.usage` resolves `undefined` (NULL token counts). Don't remove during refactor.
   - **Tool-call-only turns may emit a leading `\n` text-delta.** `MessageList.flatten`'s `hasText` and `MessageBubble`'s `hasContent` both `.trim()` before the length check, else whitespace-only content renders an empty bubble + ActionRow between tool calls. `buildMessagesPayload` also skips `status='failed'` and complete-but-empty assistant rows (avoids "Cannot have 2 or more assistant messages at the end of the list" upstream rejection after cap-hit + Continue).
+- **`services/inference/tool-shim.ts`** — Recovers structured tool calls from plain-text model output. Some models (notably Qwen) emit `<tool_call><name>...</name><arguments>...</arguments></tool_call>` inline text instead of structured JSON. `extractToolCalls(text)` parses both XML and JSON inline formats. `hasToolCallMarkup(text)` is a fast pre-check. Used as a fallback in the stream phase when structured `tool_calls` parse fails. Does NOT require `FAST_MODEL` — operates on the existing turn's output text.
+- **`services/inference/loop-detectors.ts`** — Six detectors that catch repetitive model behavior: `detectContentRepeat` (same content N times), `detectToolLoop` (same tool called consecutively). `detectDoomLoop` combines both. These are additive to the existing `sentinels.ts` doom-loop detection.
 - **AI SDK ModelMessage conversion** (`toModelMessages` in stream-phase.ts). Tool messages need a `toolName` for `ToolResultPart`; BooCode's OpenAI-shape history lacks it, so a forward-scan builds a `tool_call_id → toolName` map from prior assistant `tool_calls`. Tool outputs wrapped as `{ type: 'json' | 'text', value }` (v6 `ToolResultOutput`). Reasoning emits a `ReasoningPart` first in the content array.
 - **`experimental_repairToolCall`** wired into `streamText` to keep the stream alive when qwen3.6 emits malformed tool args. Pass-through: logs the bad call, returns it unmodified; `executeToolPhase`'s zod-reject path routes it back to the model next turn.
 - **`chat_status` frame** (via `broker.publishUser`) — `status: 'streaming' | 'tool_running' | 'waiting_for_input' | 'idle' | 'error'`. Frontend `useChatStatus` derives `idle_warm` (<30s since idle) vs `idle_cold`. `ChatThroughput` renders beside `StatusDot` only when streaming/tool_running, fed by 500ms-throttled `'usage'` frames (`completion_tokens` + `ctx_used` + `ctx_max`). `POST /api/chats/:id/discard_stale` marks a stuck-streaming row `failed` when the frontend's 60s no-token timer gives up.

apps/server/src/routes/projects.ts

@@ -1,6 +1,6 @@
 import type { FastifyInstance } from 'fastify';
 import { z } from 'zod';
-import { realpath, stat, readdir, access } from 'node:fs/promises';
+import { realpath, stat, readdir, access, writeFile, rename } from 'node:fs/promises';
 import { basename, resolve, sep } from 'node:path';
 import type { Sql } from '../db.js';
 import type { Config } from '../config.js';
@@ -473,7 +473,7 @@ export function registerProjectRoutes(
   // Always includes auto_mode (the dirty-state-derived mode) so the client can
   // show a suggestion when a pinned mode diverges from what would be auto-selected.
   // Returns { git_repo: false } when the path is not a git repository.
-  app.get<{ Params: { id: string }; Querystring: { mode?: string } }>(
+  app.get<{ Params: { id: string }; Querystring: { mode?: string; whitespace?: string } }>(
     '/api/projects/:id/git/diff',
     async (req, reply) => {
       const { id } = req.params;
@@ -504,7 +504,8 @@ export function registerProjectRoutes(
         rawMode === 'uncommitted' ? 'uncommitted' :
         auto_mode; // no mode param → auto-select (FIX 1)
 
-      const result = await getGitDiff(projectRoot, mode);
+      const ignoreWhitespace = req.query.whitespace === '1';
+      const result = await getGitDiff(projectRoot, mode, ignoreWhitespace);
       if (result === null) {
         return { git_repo: false, mode, auto_mode, base_label: null, in_progress_op: null, files: [] };
       }
@@ -541,6 +542,11 @@ export function registerProjectRoutes(
     ).min(1),
   });
 
+  const WriteFileBody = z.object({
+    path: z.string().min(1),
+    content: z.string(),
+  });
+
   // POST /api/projects/:id/git/stage — stage whole files
   app.post<{ Params: { id: string } }>(
     '/api/projects/:id/git/stage',
@@ -637,6 +643,38 @@ export function registerProjectRoutes(
     },
   );
 
+  // POST /api/projects/:id/write_file — write a file atomically
+  app.post<{ Params: { id: string } }>(
+    '/api/projects/:id/write_file',
+    async (req, reply) => {
+      const body = WriteFileBody.safeParse(req.body);
+      if (!body.success) { reply.code(400); return { error: body.error.message }; }
+      const { id } = req.params;
+      const projectPath = await selectProjectPath(sql, id);
+      if (!projectPath) { reply.code(404); return { error: 'not found' }; }
+      let root: string;
+      try { root = await resolveProjectRoot(projectPath); }
+      catch (err) { if (err instanceof PathScopeError) { reply.code(404); return { error: (err as Error).message }; } throw err; }
+      const target = body.data.path.startsWith('/') ? body.data.path : resolve(root, body.data.path);
+      // Validate path stays within project root
+      const realTarget = await realpath(target).catch(() => target);
+      if (!realTarget.startsWith(root + sep) && realTarget !== root) {
+        reply.code(403);
+        return { error: 'path escapes project root' };
+      }
+      const tmp = target + '.tmp';
+      try {
+        await writeFile(tmp, body.data.content, 'utf-8');
+        await rename(tmp, target);
+        return { ok: true };
+      } catch (err) {
+        // Clean up tmp on failure
+        await access(tmp).then(() => rename(tmp, target + '.bak').catch(() => {})).catch(() => {});
+        throw err;
+      }
+    },
+  );
+
   // GET /api/projects/:id/files
   app.get<{ Params: { id: string } }>(
     '/api/projects/:id/files',

apps/server/src/schema.sql

@@ -372,13 +372,12 @@ ALTER TABLE messages ADD COLUMN IF NOT EXISTS tail_start_id UUID REFERENCES mess
 ALTER TABLE chats ADD COLUMN IF NOT EXISTS needs_compaction BOOLEAN NOT NULL DEFAULT FALSE;
 CREATE INDEX IF NOT EXISTS idx_messages_chat_compacted ON messages (chat_id, compacted_at);
 
--- tasks table (provider dispatch, arena)
+-- tasks table (provider dispatch)
 CREATE TABLE IF NOT EXISTS tasks (
   id                UUID PRIMARY KEY DEFAULT gen_random_uuid(),
   project_id        UUID NOT NULL REFERENCES projects(id) ON DELETE CASCADE,
   session_id        UUID REFERENCES sessions(id) ON DELETE CASCADE,
   parent_task_id    UUID REFERENCES tasks(id),
-  arena_id          UUID,
   state             TEXT NOT NULL DEFAULT 'pending'
                     CHECK (state IN ('pending','running','completed','failed','blocked','cancelled')),
   input             TEXT NOT NULL,
@@ -405,3 +404,6 @@ DO $$ BEGIN
       FOREIGN KEY (session_id) REFERENCES sessions(id) ON DELETE CASCADE;
   END IF;
 END $$;
+
+-- Remove the v2.0.5 arena_id column (replaced by the new Arena feature).
+ALTER TABLE tasks DROP COLUMN IF EXISTS arena_id;

apps/server/src/services/git_diff.ts

@@ -271,7 +271,9 @@ function buildNumstatMap(
 async function getUncommittedDiff(
   gitRoot: string,
   inProgress: string | null,
+  ignoreWhitespace = false,
 ): Promise<GitDiffResult> {
+  const ws = ignoreWhitespace ? ['-w'] : [];
   const hasCommits = (await runGit(['rev-parse', '--verify', 'HEAD'], gitRoot)) !== null;
 
   const [nameStatusOut, cachedNameStatusOut, untrackedOut, numstatOut, diffOut, cachedDiffOut] =
@@ -284,10 +286,10 @@ async function getUncommittedDiff(
         : runGit(['diff', '--cached', '--name-status'], gitRoot),
       runGit(['ls-files', '--others', '--exclude-standard'], gitRoot),
       hasCommits ? runGit(['diff', '--numstat', 'HEAD'], gitRoot) : Promise.resolve(''),
-      hasCommits ? runGit(['diff', 'HEAD'], gitRoot) : Promise.resolve(''),
+      hasCommits ? runGit(['diff', ...ws, 'HEAD'], gitRoot) : Promise.resolve(''),
       hasCommits
-        ? runGit(['diff', '--cached', 'HEAD'], gitRoot)
+        ? runGit(['diff', ...ws, '--cached', 'HEAD'], gitRoot)
-        : runGit(['diff', '--cached'], gitRoot),
+        : runGit(['diff', ...ws, '--cached'], gitRoot),
     ]);
 
   const allChanged = parseNameStatus(nameStatusOut ?? '');
@@ -347,11 +349,13 @@ async function getCommittedDiff(
   base: string,
   label: string,
   inProgress: string | null,
+  ignoreWhitespace = false,
 ): Promise<GitDiffResult> {
+  const ws = ignoreWhitespace ? ['-w'] : [];
   const [nameStatusOut, numstatOut, diffOut] = await Promise.all([
     runGit(['diff', '--name-status', base, 'HEAD'], gitRoot),
     runGit(['diff', '--numstat', base, 'HEAD'], gitRoot),
-    runGit(['diff', base, 'HEAD'], gitRoot),
+    runGit(['diff', ...ws, base, 'HEAD'], gitRoot),
   ]);
 
   const allChanged = parseNameStatus(nameStatusOut ?? '');
@@ -383,23 +387,23 @@ async function getCommittedDiff(
  * the directory is not a git repository. On a null committed-mode base, falls
  * back to uncommitted and labels the result accordingly.
  */
-export async function getGitDiff(cwd: string, mode: GitDiffMode): Promise<GitDiffResult | null> {
+export async function getGitDiff(cwd: string, mode: GitDiffMode, ignoreWhitespace?: boolean): Promise<GitDiffResult | null> {
   const gitRoot = await resolveGitRoot(cwd);
   if (!gitRoot) return null;
 
   const inProgress = await detectInProgress(gitRoot);
 
   if (mode === 'uncommitted') {
-    return getUncommittedDiff(gitRoot, inProgress);
+    return getUncommittedDiff(gitRoot, inProgress, ignoreWhitespace ?? false);
   }
 
   const { base, label } = await resolveCommittedBase(gitRoot);
   if (!base) {
     // Fall back to uncommitted with a descriptive label
-    const result = await getUncommittedDiff(gitRoot, inProgress);
+    const result = await getUncommittedDiff(gitRoot, inProgress, ignoreWhitespace ?? false);
     return { ...result, base_label: label };
   }
-  return getCommittedDiff(gitRoot, base, label, inProgress);
+  return getCommittedDiff(gitRoot, base, label, inProgress, ignoreWhitespace ?? false);
 }
 
 // ── Phase 2: Write helpers ─────────────────────────────────────────────────

apps/server/src/services/inference/dcp/__tests__/deduplication.test.ts

@@ -0,0 +1,33 @@
+import { describe, it, expect } from 'vitest';
+import { deduplicate } from '../strategies/deduplication.js';
+import type { DcpMessage } from '../messages.js';
+
+describe('deduplicate', () => {
+  it('removes consecutive identical tool_call+tool_result pairs', () => {
+    const messages: DcpMessage[] = [
+      { role: 'user', content: 'search for x' },
+      { role: 'assistant', content: '', tool_calls: [{ id: '1', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: 'result1', tool_call_id: '1' },
+      // Duplicate pair
+      { role: 'assistant', content: '', tool_calls: [{ id: '2', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: 'result1', tool_call_id: '2' },
+    ];
+
+    const { messages: result, stats } = deduplicate(messages);
+    expect(result).toHaveLength(3); // user + first pair
+    expect(stats.removedCount).toBe(2);
+  });
+
+  it('preserves non-duplicate content', () => {
+    const messages: DcpMessage[] = [
+      { role: 'assistant', content: '', tool_calls: [{ id: '1', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: 'result1', tool_call_id: '1' },
+      { role: 'assistant', content: '', tool_calls: [{ id: '2', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: 'result2', tool_call_id: '2' }, // Different result
+    ];
+
+    const { messages: result, stats } = deduplicate(messages);
+    expect(result).toHaveLength(4);
+    expect(stats.removedCount).toBe(0);
+  });
+});

apps/server/src/services/inference/dcp/__tests__/messages.test.ts

@@ -0,0 +1,22 @@
+import { describe, it, expect } from 'vitest';
+import { toDcpMessages, fromDcpMessages } from '../messages.js';
+
+describe('toDcpMessages', () => {
+  it('converts user messages', () => {
+    const result = toDcpMessages([{ role: 'user', content: 'hello' }]);
+    expect(result[0].role).toBe('user');
+    expect(result[0].content).toBe('hello');
+  });
+
+  it('marks Error: content as isError', () => {
+    const result = toDcpMessages([{ role: 'tool', content: 'Error: file not found', tool_call_id: '1' }]);
+    expect(result[0].isError).toBe(true);
+  });
+});
+
+describe('fromDcpMessages', () => {
+  it('round-trips messages', () => {
+    const original = [{ role: 'user', content: 'hello' }];
+    expect(fromDcpMessages(toDcpMessages(original))).toEqual(original);
+  });
+});

apps/server/src/services/inference/dcp/__tests__/purge-errors.test.ts

@@ -0,0 +1,33 @@
+import { describe, it, expect } from 'vitest';
+import { purgeErrors } from '../strategies/purge-errors.js';
+import type { DcpMessage } from '../messages.js';
+
+describe('purgeErrors', () => {
+  it('removes tool results where content starts with Error:', () => {
+    const messages: DcpMessage[] = [
+      { role: 'tool', content: 'Error: file not found', tool_call_id: '1' },
+      { role: 'tool', content: '{"files":[]}', tool_call_id: '2' },
+    ];
+    const { messages: result, stats } = purgeErrors(messages);
+    expect(result).toHaveLength(1);
+    expect(stats.removedCount).toBe(1);
+  });
+
+  it('removes empty tool results', () => {
+    const messages: DcpMessage[] = [
+      { role: 'tool', content: '', tool_call_id: '1' },
+    ];
+    const { messages: result, stats } = purgeErrors(messages);
+    expect(result).toHaveLength(0);
+    expect(stats.removedCount).toBe(1);
+  });
+
+  it('preserves valid tool results', () => {
+    const messages: DcpMessage[] = [
+      { role: 'tool', content: '{"files":["a.ts"]}', tool_call_id: '1' },
+    ];
+    const { messages: result, stats } = purgeErrors(messages);
+    expect(result).toHaveLength(1);
+    expect(stats.removedCount).toBe(0);
+  });
+});

apps/server/src/services/inference/dcp/__tests__/transform.test.ts

@@ -0,0 +1,25 @@
+import { describe, it, expect } from 'vitest';
+import { transformMessages } from '../transform.js';
+import type { DcpMessage } from '../messages.js';
+
+describe('transformMessages', () => {
+  it('applies dedup then purge in order', () => {
+    const input: DcpMessage[] = [
+      { role: 'user', content: 'hello' },
+      { role: 'assistant', content: '', tool_calls: [{ id: '1', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: 'result', tool_call_id: '1' },
+      { role: 'assistant', content: '', tool_calls: [{ id: '2', name: 'grep', arguments: '{}' }] },
+      { role: 'tool', content: 'result', tool_call_id: '2' }, // Dup
+    ];
+
+    const { messages, stats } = transformMessages('test-chat', input);
+    expect(stats.removedCount).toBeGreaterThan(0);
+    expect(messages.length).toBeLessThan(input.length);
+  });
+
+  it('handles empty input', () => {
+    const { messages, stats } = transformMessages('empty', []);
+    expect(messages).toHaveLength(0);
+    expect(stats.removedCount).toBe(0);
+  });
+});

apps/server/src/services/inference/dcp/index.ts

@@ -0,0 +1,4 @@
+export { transformMessages } from './transform.js';
+export type { DcpMessage } from './messages.js';
+export { toDcpMessages, fromDcpMessages } from './messages.js';
+export { getDcpState, clearDcpState } from './state.js';

apps/server/src/services/inference/dcp/messages.ts

@@ -0,0 +1,34 @@
+// DCP message shape adapter.
+// Converts between BooCode MessagePart[] and the DCP internal shape.
+// Clean-room implementation — no AGPL source copied.
+
+export interface DcpMessage {
+  role: 'user' | 'assistant' | 'tool';
+  content: string;
+  tool_call_id?: string;
+  tool_calls?: Array<{ id: string; name: string; arguments: string }>;
+  isError?: boolean;
+}
+
+export function toDcpMessages(parts: any[]): DcpMessage[] {
+  return parts.map((p: any) => {
+    const msg: DcpMessage = { role: p.role, content: p.content ?? '' };
+    if (p.tool_call_id) msg.tool_call_id = p.tool_call_id;
+    if (p.tool_calls) msg.tool_calls = p.tool_calls;
+    if (p.isError) msg.isError = true;
+    if (p.role === 'tool' && p.content && p.content.startsWith('Error:')) {
+      msg.isError = true;
+    }
+    return msg;
+  });
+}
+
+export function fromDcpMessages(msgs: DcpMessage[]): any[] {
+  return msgs.map((m) => ({
+    role: m.role,
+    content: m.content,
+    ...(m.tool_call_id ? { tool_call_id: m.tool_call_id } : {}),
+    ...(m.tool_calls ? { tool_calls: m.tool_calls } : {}),
+    ...(m.isError ? { isError: true } : {}),
+  }));
+}

apps/server/src/services/inference/dcp/state.ts

@@ -0,0 +1,27 @@
+// Per-chat session state for DCP.
+// Tracks last transform timestamp and message count to avoid re-processing.
+
+interface ChatDcpState {
+  lastTransformAt: number;
+  lastMessageCount: number;
+}
+
+const chatStates = new Map<string, ChatDcpState>();
+
+export function getDcpState(chatId: string): ChatDcpState | undefined {
+  return chatStates.get(chatId);
+}
+
+export function setDcpState(chatId: string, messageCount: number): void {
+  chatStates.set(chatId, { lastTransformAt: Date.now(), lastMessageCount: messageCount });
+}
+
+export function clearDcpState(chatId: string): void {
+  chatStates.delete(chatId);
+}
+
+export function shouldTransform(chatId: string, messageCount: number): boolean {
+  const state = chatStates.get(chatId);
+  if (!state) return true;
+  return state.lastMessageCount !== messageCount;
+}

apps/server/src/services/inference/dcp/strategies/deduplication.ts

@@ -0,0 +1,50 @@
+import type { DcpMessage } from '../messages.js';
+
+export function deduplicate(messages: DcpMessage[]): { messages: DcpMessage[]; stats: { removedCount: number; freedTokens: number } } {
+  const result: DcpMessage[] = [];
+  let removedCount = 0;
+  let freedTokens = 0;
+  let i = 0;
+
+  while (i < messages.length) {
+    const current: DcpMessage = messages[i]!;
+    const next = messages[i + 1];
+
+    if (
+      current.role === 'assistant' &&
+      current.tool_calls &&
+      next &&
+      next.role === 'tool' &&
+      next.tool_call_id === current.tool_calls[0]?.id
+    ) {
+      const nextNext = messages[i + 2];
+      const nextNextNext = messages[i + 3];
+
+      if (
+        nextNext &&
+        nextNext.role === 'assistant' &&
+        nextNext.tool_calls &&
+        nextNextNext &&
+        nextNextNext.role === 'tool' &&
+        nextNextNext.tool_call_id === nextNext.tool_calls[0]?.id &&
+        nextNext.tool_calls[0]?.name === current.tool_calls[0]?.name &&
+        nextNext.tool_calls[0]?.arguments === current.tool_calls[0]?.arguments &&
+        nextNextNext.content === next.content
+      ) {
+        result.push(current, next);
+        i += 4;
+        removedCount += 2;
+        freedTokens += Math.ceil(nextNext.content.length / 4);
+        freedTokens += Math.ceil(current.content.length / 4);
+      } else {
+        result.push(current);
+        i++;
+      }
+    } else {
+      result.push(current);
+      i++;
+    }
+  }
+
+  return { messages: result, stats: { removedCount, freedTokens } };
+}

apps/server/src/services/inference/dcp/strategies/purge-errors.ts

@@ -0,0 +1,34 @@
+// Purge-errors strategy — removes failed/empty tool_result entries.
+// Clean-room implementation.
+
+import type { DcpMessage } from '../messages.js';
+
+const ERROR_PREFIXES = ['Error:', 'error:', 'Error: '];
+const DEFAULT_WINDOW = 5;
+
+export function purgeErrors(
+  messages: DcpMessage[],
+  windowSize: number = DEFAULT_WINDOW,
+): { messages: DcpMessage[]; stats: { removedCount: number; freedTokens: number } } {
+  const result: DcpMessage[] = [];
+  let removedCount = 0;
+  let freedTokens = 0;
+
+  for (const msg of messages) {
+    if (msg.role === 'tool') {
+      const shouldRemove =
+        msg.isError ||
+        ERROR_PREFIXES.some((p) => msg.content.startsWith(p)) ||
+        msg.content.trim() === '';
+
+      if (shouldRemove) {
+        removedCount++;
+        freedTokens += Math.ceil(msg.content.length / 4);
+        continue; // Skip this message
+      }
+    }
+    result.push(msg);
+  }
+
+  return { messages: result, stats: { removedCount, freedTokens } };
+}

apps/server/src/services/inference/dcp/transform.ts

@@ -0,0 +1,52 @@
+// Transform orchestrator — runs DCP strategies in sequence.
+// Clean-room implementation.
+
+import type { DcpMessage } from './messages.js';
+import { deduplicate } from './strategies/deduplication.js';
+import { purgeErrors } from './strategies/purge-errors.js';
+import { getDcpState, setDcpState, shouldTransform } from './state.js';
+
+export interface TransformStats {
+  removedCount: number;
+  freedTokens: number;
+  dedupRemoved: number;
+  purgeRemoved: number;
+}
+
+export interface TransformResult {
+  messages: DcpMessage[];
+  stats: TransformStats;
+}
+
+export function transformMessages(chatId: string, messages: DcpMessage[]): TransformResult {
+  if (!shouldTransform(chatId, messages.length)) {
+    return { messages, stats: { removedCount: 0, freedTokens: 0, dedupRemoved: 0, purgeRemoved: 0 } };
+  }
+
+  let m = messages;
+
+  // Step 1: Deduplicate
+  const dedupResult = deduplicate(m);
+  m = dedupResult.messages;
+  const dedupRemoved = dedupResult.stats.removedCount;
+
+  // Step 2: Purge errors
+  const purgeResult = purgeErrors(m);
+  m = purgeResult.messages;
+  const purgeRemoved = purgeResult.stats.removedCount;
+
+  const totalRemoved = dedupRemoved + purgeRemoved;
+  const totalFreed = dedupResult.stats.freedTokens + purgeResult.stats.freedTokens;
+
+  setDcpState(chatId, messages.length);
+
+  return {
+    messages: m,
+    stats: {
+      removedCount: totalRemoved,
+      freedTokens: totalFreed,
+      dedupRemoved,
+      purgeRemoved,
+    },
+  };
+}

apps/server/src/services/inference/loop-detectors.ts

@@ -0,0 +1,68 @@
+// Loop detectors — detects repetitive patterns in assistant output
+// that indicate a model is stuck in a loop.
+
+export interface LoopDetectionResult {
+  isLoop: boolean;
+  reason?: string;
+  confidence: number; // 0-1
+}
+
+const REPEATED_PHRASE_MIN_COUNT = 4;
+const REPEATED_TOOL_MIN_COUNT = 3;
+
+export function detectContentRepeat(messages: string[]): LoopDetectionResult {
+  if (messages.length < REPEATED_PHRASE_MIN_COUNT) {
+    return { isLoop: false, confidence: 0 };
+  }
+
+  const recent = messages.slice(-REPEATED_PHRASE_MIN_COUNT);
+  const unique = new Set(recent);
+
+  if (unique.size === 1) {
+    return {
+      isLoop: true,
+      reason: `Same content repeated ${REPEATED_PHRASE_MIN_COUNT} times`,
+      confidence: 0.9,
+    };
+  }
+
+  if (unique.size <= 2 && recent.length >= 4) {
+    return {
+      isLoop: true,
+      reason: 'Content oscillating between two variants',
+      confidence: 0.7,
+    };
+  }
+
+  return { isLoop: false, confidence: 0 };
+}
+
+export function detectToolLoop(toolNames: string[]): LoopDetectionResult {
+  if (toolNames.length < REPEATED_TOOL_MIN_COUNT) return { isLoop: false, confidence: 0 };
+
+  const recent = toolNames.slice(-REPEATED_TOOL_MIN_COUNT);
+  const unique = new Set(recent);
+
+  if (unique.size === 1) {
+    return {
+      isLoop: true,
+      reason: `Same tool "${recent[0]}" called ${REPEATED_TOOL_MIN_COUNT} times consecutively`,
+      confidence: 0.85,
+    };
+  }
+
+  return { isLoop: false, confidence: 0 };
+}
+
+export function detectDoomLoop(
+  messages: string[],
+  toolNames: string[],
+): LoopDetectionResult {
+  const contentResult = detectContentRepeat(messages);
+  if (contentResult.isLoop) return contentResult;
+
+  const toolResult = detectToolLoop(toolNames);
+  if (toolResult.isLoop) return toolResult;
+
+  return { isLoop: false, confidence: 0 };
+}

apps/server/src/services/inference/tool-shim.ts

@@ -0,0 +1,45 @@
+// ToolShim — recovers structured tool calls from plain-text model output.
+// When the model emits tool calls as plain text instead of structured JSON,
+// this shim attempts to parse and recover them.
+
+export interface ParsedToolCall {
+  id: string;
+  name: string;
+  arguments: string;
+}
+
+const TOOL_CALL_PATTERN = /<tool_call>\s*<name>(.+?)<\/name>\s*<arguments>(.+?)<\/arguments>\s*<\/tool_call>/gs;
+const JSON_TOOL_PATTERN = /\{\s*"name":\s*"([^"]+)",\s*"arguments":\s*({.+?})\s*\}/gs;
+
+export function extractToolCalls(text: string): ParsedToolCall[] {
+  const calls: ParsedToolCall[] = [];
+  let match: RegExpExecArray | null;
+
+  // Try XML-style tool calls (common in Qwen output)
+  const xmlRegex = new RegExp(TOOL_CALL_PATTERN);
+  while ((match = xmlRegex.exec(text)) !== null) {
+    calls.push({
+      id: `call_${calls.length}`,
+      name: match[1]!.trim(),
+      arguments: match[2]!.trim(),
+    });
+  }
+
+  if (calls.length > 0) return calls;
+
+  // Try JSON-style tool calls
+  const jsonRegex = new RegExp(JSON_TOOL_PATTERN);
+  while ((match = jsonRegex.exec(text)) !== null) {
+    calls.push({
+      id: `call_${calls.length}`,
+      name: match[1]!.trim(),
+      arguments: match[2]!.trim(),
+    });
+  }
+
+  return calls;
+}
+
+export function hasToolCallMarkup(text: string): boolean {
+  return TOOL_CALL_PATTERN.test(text) || JSON_TOOL_PATTERN.test(text);
+}

apps/server/src/services/inference/turn.ts

@@ -21,6 +21,7 @@ import {
   buildMessagesPayload,
   loadContext,
 } from './payload.js';
+import { toDcpMessages, transformMessages, fromDcpMessages } from './dcp/index.js';
 import {
   finalizeCompletion,
   finalizeEmpty,
@@ -156,9 +157,20 @@ export async function runAssistantTurn(
       ctx.log.warn({ sessionId }, 'inference: session or project missing mid-loop');
       break;
     }
-    const { session: iterSession, project: iterProject, history } = loaded;
+    let { session: iterSession, project: iterProject, history } = loaded;
     const projectRoot = await resolveProjectRoot(iterProject.path);
 
+    try {
+      const dcpMsgs = toDcpMessages(history);
+      const { messages: pruned, stats } = transformMessages(chatId, dcpMsgs);
+      if (stats.removedCount > 0) {
+        ctx.log.info({ chatId, ...stats }, 'dcp: transform removed messages');
+        history = fromDcpMessages(pruned) as typeof history;
+      }
+    } catch (err) {
+      ctx.log.warn({ err: err instanceof Error ? err.message : String(err), chatId }, 'dcp: transform skipped');
+    }
+
     // v1.14.0: log step boundary for instrumentation. step_start parts are in
     // the schema CHECK but not emitted here — writing to the assistant message
     // before the stream phase creates a sequence-0 collision with

apps/server/src/services/inference/types.ts

@@ -44,7 +44,11 @@ export interface InferenceFrame {
     | 'chat_renamed'
     | 'error'
     | 'flow_run_started'
-    | 'flow_run_step_updated';
+    | 'flow_run_step_updated'
+    // arena frames
+    | 'battle_started'
+    | 'contestant_updated'
+    | 'battle_updated';
   message_id?: string;
   message_ids?: string[];
   chat_id?: string;
@@ -84,6 +88,19 @@ export interface InferenceFrame {
   status?: string;
   run_status?: 'running' | 'completed' | 'failed' | 'cancelled';
   report?: string;
+  // arena frames
+  battle_id?: string;
+  battle_type?: 'coding' | 'qa';
+  prompt?: string;
+  contestants?: Array<{ id: string; identity: string; model: string; lane: 'local' | 'cloud' }>;
+  contestant_id?: string;
+  battle_status?: 'pending' | 'running' | 'completed' | 'failed' | 'cancelled';
+  duration_ms?: number;
+  tokens_per_sec?: number;
+  winner_contestant_id?: string | null;
+  analysis_ready?: boolean;
+  cross_exam_id?: string;
+  delta?: string;
 }
 
 export type FramePublisher = (sessionId: string, frame: InferenceFrame) => void;

apps/server/src/services/memory/__tests__/bm25.test.ts

@@ -0,0 +1,37 @@
+import { describe, it, expect } from 'vitest';
+import { Bm25Ranker } from '../bm25.js';
+
+describe('Bm25Ranker', () => {
+  it('scores documents by term frequency', () => {
+    const ranker = new Bm25Ranker();
+    ranker.fit(['the cat sat on the mat', 'the dog chased the cat', 'the bird flew over the mat']);
+    const results = ranker.rank('cat mat');
+    expect(results.length).toBeGreaterThan(0);
+    expect(results[0]!.score).toBeGreaterThan(0);
+  });
+
+  it('returns empty for no matches', () => {
+    const ranker = new Bm25Ranker();
+    ranker.fit(['aaa bbb', 'ccc ddd']);
+    const results = ranker.rank('zzz');
+    expect(results).toHaveLength(0);
+  });
+
+  it('handles single document corpus', () => {
+    const ranker = new Bm25Ranker();
+    ranker.fit(['only document here']);
+    const results = ranker.rank('document');
+    expect(results).toHaveLength(1);
+  });
+
+  it('ranks relevant docs higher', () => {
+    const ranker = new Bm25Ranker();
+    ranker.fit([
+      'javascript is a programming language',
+      'python is also a programming language',
+      'the weather is nice today',
+    ]);
+    const results = ranker.rank('javascript programming');
+    expect(results[0]!.index).toBe(0);
+  });
+});

apps/server/src/services/memory/__tests__/entries.test.ts

@@ -0,0 +1,31 @@
+import { describe, it, expect } from 'vitest';
+import { parseMemoryEntries } from '../entries.js';
+
+describe('parseMemoryEntries', () => {
+  it('parses a single entry with tags', () => {
+    const md = '## project: Indentation\n> tags: style\n\nUse two-space indentation\n';
+    const entries = parseMemoryEntries('style.md', md);
+    expect(entries).toHaveLength(1);
+    expect(entries[0].title).toBe('Indentation');
+    expect(entries[0].topic).toBe('project');
+    expect(entries[0].tags).toEqual(['style']);
+    expect(entries[0].content).toContain('two-space');
+  });
+
+  it('parses multiple entries', () => {
+    const md = [
+      '## project: Style',
+      '',
+      'Use tab indentation',
+      '',
+      '## user: Preference',
+      '',
+      'Prefer pnpm',
+      '',
+    ].join('\n');
+    const entries = parseMemoryEntries('mem.md', md);
+    expect(entries).toHaveLength(2);
+    expect(entries[0].topic).toBe('project');
+    expect(entries[1].topic).toBe('user');
+  });
+});

apps/server/src/services/memory/__tests__/paths.test.ts

@@ -0,0 +1,14 @@
+import { describe, it, expect } from 'vitest';
+import { getMemoryRoot, getTopicDir } from '../paths.js';
+
+describe('getMemoryRoot', () => {
+  it('returns .boocode/memory under project root', () => {
+    expect(getMemoryRoot('/proj')).toBe('/proj/.boocode/memory');
+  });
+});
+
+describe('getTopicDir', () => {
+  it('returns project/ under memory root', () => {
+    expect(getTopicDir('/r/.boocode/memory', 'project')).toBe('/r/.boocode/memory/project');
+  });
+});

apps/server/src/services/memory/__tests__/prompt.test.ts

@@ -0,0 +1,15 @@
+import { describe, it, expect } from 'vitest';
+import { formatMemoryBlock } from '../prompt.js';
+
+describe('formatMemoryBlock', () => {
+  it('wraps entries in boocode-memory tags', () => {
+    const block = formatMemoryBlock(['Use pnpm', 'Tests in vitest']);
+    expect(block).toContain('<boocode-memory>');
+    expect(block).toContain('Use pnpm');
+    expect(block).toContain('</boocode-memory>');
+  });
+
+  it('returns empty string for no entries', () => {
+    expect(formatMemoryBlock([])).toBe('');
+  });
+});

apps/server/src/services/memory/__tests__/recall.test.ts

@@ -0,0 +1,27 @@
+import { describe, it, expect } from 'vitest';
+import { rankByRelevance } from '../recall.js';
+import type { MemoryEntry } from '../entries.js';
+
+describe('rankByRelevance', () => {
+  it('returns entries matching query keywords', () => {
+    const entries: MemoryEntry[] = [
+      { id: '1', topic: 'project', title: 'Style', content: 'Use two-space indentation', tags: ['style'] },
+      { id: '2', topic: 'project', title: 'Tests', content: 'Use vitest for testing', tags: ['testing'] },
+    ];
+    const result = rankByRelevance('what indentation?', entries);
+    expect(result).toHaveLength(1);
+    expect(result[0].title).toBe('Style');
+  });
+});
+
+describe('rankByHybrid', () => {
+  it('falls back to BM25 when embeddings unavailable', async () => {
+    const entries: MemoryEntry[] = [
+      { id: '1', topic: 'project', title: 'Style', content: 'Use two-space indentation', tags: ['style'] },
+      { id: '2', topic: 'project', title: 'Tests', content: 'Use vitest for testing', tags: ['testing'] },
+    ];
+    const { rankByHybrid } = await import('../recall.js');
+    const result = await rankByHybrid('indentation style', entries);
+    expect(result.length).toBeGreaterThan(0);
+  });
+});

apps/server/src/services/memory/bm25.ts

@@ -0,0 +1,67 @@
+// BM25 ranker — pure Okapi BM25 scoring. No external deps.
+
+interface Bm25Config {
+  k1?: number;
+  b?: number;
+}
+
+export class Bm25Ranker {
+  private k1: number;
+  private b: number;
+  private corpus: string[];
+  private avgDocLen: number;
+  private idfCache: Map<string, number>;
+  private docCount: number;
+
+  constructor(config?: Bm25Config) {
+    this.k1 = config?.k1 ?? 1.5;
+    this.b = config?.b ?? 0.75;
+    this.corpus = [];
+    this.avgDocLen = 0;
+    this.idfCache = new Map();
+    this.docCount = 0;
+  }
+
+  fit(docs: string[]): void {
+    this.corpus = docs;
+    this.docCount = docs.length;
+    const lengths = docs.map((d) => d.split(/\s+/).length);
+    this.avgDocLen = lengths.reduce((a, b) => a + b, 0) / lengths.length;
+    this.idfCache.clear();
+  }
+
+  private tokenize(text: string): string[] {
+    return text.toLowerCase().split(/\s+/).filter((t) => t.length > 0);
+  }
+
+  private idf(term: string): number {
+    const cached = this.idfCache.get(term);
+    if (cached !== undefined) return cached;
+    const docsWithTerm = this.corpus.filter((d) => this.tokenize(d).includes(term)).length;
+    const idf = Math.log(1 + (this.docCount - docsWithTerm + 0.5) / (docsWithTerm + 0.5));
+    this.idfCache.set(term, idf);
+    return idf;
+  }
+
+  score(query: string, docIndex: number): number {
+    if (docIndex < 0 || docIndex >= this.corpus.length) return 0;
+    const doc = this.corpus[docIndex]!;
+    const queryTerms = this.tokenize(query);
+    const docTokens = this.tokenize(doc);
+    const docLen = docTokens.length;
+
+    let total = 0;
+    for (const term of queryTerms) {
+      const tf = docTokens.filter((t) => t === term).length;
+      if (tf === 0) continue;
+      const idfVal = this.idf(term);
+      total += idfVal * ((tf * (this.k1 + 1)) / (tf + this.k1 * (1 - this.b + this.b * docLen / this.avgDocLen)));
+    }
+    return total;
+  }
+
+  rank(query: string, topN: number = 10): Array<{ index: number; score: number }> {
+    const scores = this.corpus.map((_, i) => ({ index: i, score: this.score(query, i) }));
+    return scores.sort((a, b) => b.score - a.score).slice(0, topN).filter((s) => s.score > 0);
+  }
+}

apps/server/src/services/memory/embeddings.ts

@@ -0,0 +1,55 @@
+// Embedding module — ONNX-based local embeddings.
+// Falls back gracefully when the model file is not available.
+
+let model: any = null;
+let ortModule: any = null;
+
+export function isEmbeddingAvailable(): boolean {
+  return model !== null;
+}
+
+// eslint-disable-next-line @typescript-eslint/no-require-imports
+const dynamicRequire = typeof require !== 'undefined' ? require : null;
+
+export async function initEmbeddings(modelPath?: string): Promise<boolean> {
+  try {
+    if (dynamicRequire) {
+      try { ortModule = dynamicRequire('onnxruntime-node'); } catch { ortModule = null; }
+    }
+    if (!ortModule) {
+      try { ortModule = await import('onnxruntime-node' as any); } catch { ortModule = null; }
+    }
+    if (!ortModule) return false;
+    const path = modelPath ?? process.env['EMBEDDING_MODEL_PATH'] ?? '';
+    if (!path) return false;
+    model = await ortModule.InferenceSession.create(path);
+    return true;
+  } catch {
+    model = null;
+    return false;
+  }
+}
+
+export async function embed(texts: string[]): Promise<number[][] | null> {
+  if (!model) return null;
+  try {
+    // eslint-disable-next-line @typescript-eslint/no-unnecessary-condition
+    const ort: { Tensor: new (...args: unknown[]) => unknown } | null = ortModule || null;
+    if (!ort) return null;
+    const input = new ort.Tensor('string', texts, [texts.length]);
+    const feeds: Record<string, any> = {};
+    feeds[model.inputNames[0]] = input;
+    const results = await model.run(feeds);
+    const output = results[model.outputNames[0]];
+    if (!output || !output.data) return null;
+    const dim = output.dims?.[1] ?? 384;
+    const data = output.data as Float32Array;
+    const vectors: number[][] = [];
+    for (let i = 0; i < texts.length; i++) {
+      vectors.push(Array.from(data.slice(i * dim, (i + 1) * dim)));
+    }
+    return vectors;
+  } catch {
+    return null;
+  }
+}

apps/server/src/services/memory/entries.ts

@@ -0,0 +1,54 @@
+export interface MemoryEntry {
+  id: string;
+  topic: string;
+  title: string;
+  content: string;
+  tags: string[];
+}
+
+export function parseMemoryEntries(fileName: string, markdown: string): MemoryEntry[] {
+  const entries: MemoryEntry[] = [];
+  const lines = markdown.split('\n');
+  let currentEntry: Partial<MemoryEntry> | null = null;
+  let currentContent: string[] = [];
+
+  for (const line of lines) {
+    const headingMatch = line.match(/^##\s+(.+):\s+(.+)$/);
+    if (headingMatch && headingMatch[1] && headingMatch[2]) {
+      if (currentEntry && currentEntry.title) {
+        entries.push({
+          id: `${fileName}-${entries.length}`,
+          topic: currentEntry.topic ?? '',
+          title: currentEntry.title,
+          content: currentContent.join('\n').trim(),
+          tags: currentEntry.tags ?? [],
+        });
+      }
+      currentEntry = { topic: headingMatch[1].trim(), title: headingMatch[2].trim(), tags: [] };
+      currentContent = [];
+      continue;
+    }
+
+    const tagsMatch = line.match(/^>\s*tags:\s*(.+)$/i);
+    if (tagsMatch && tagsMatch[1] && currentEntry) {
+      currentEntry.tags = tagsMatch[1].split(',').map((t) => t.trim());
+      continue;
+    }
+
+    if (currentEntry) {
+      currentContent.push(line);
+    }
+  }
+
+  if (currentEntry && currentEntry.title) {
+    entries.push({
+      id: `${fileName}-${entries.length}`,
+      topic: currentEntry.topic ?? '',
+      title: currentEntry.title,
+      content: currentContent.join('\n').trim(),
+      tags: currentEntry.tags ?? [],
+    });
+  }
+
+  return entries;
+}

apps/server/src/services/memory/index.ts

@@ -0,0 +1,6 @@
+export { loadMemoryForSession } from './recall.js';
+export { formatMemoryBlock } from './prompt.js';
+export { scanMemoryScopes } from './scan.js';
+export { parseMemoryEntries } from './entries.js';
+export { ensureMemoryScaffold, getMemoryRoot } from './paths.js';
+export type { MemoryEntry } from './entries.js';

apps/server/src/services/memory/paths.ts

@@ -0,0 +1,17 @@
+import { join } from 'node:path';
+import { mkdir } from 'node:fs/promises';
+
+const TOPICS = ['project', 'user', 'reference'] as const;
+export type MemoryTopic = (typeof TOPICS)[number];
+
+export function getMemoryRoot(projectRoot: string): string {
+  return join(projectRoot, '.boocode', 'memory');
+}
+
+export function getTopicDir(root: string, topic: MemoryTopic): string {
+  return join(root, topic);
+}
+
+export async function ensureMemoryScaffold(root: string): Promise<void> {
+  await Promise.all(TOPICS.map((t) => mkdir(join(root, t), { recursive: true })));
+}

apps/server/src/services/memory/prompt.ts

@@ -0,0 +1,5 @@
+export function formatMemoryBlock(entries: string[]): string {
+  if (entries.length === 0) return '';
+  const body = entries.map((e) => `- ${e}`).join('\n');
+  return `<boocode-memory>\n${body}\n</boocode-memory>`;
+}

apps/server/src/services/memory/recall.ts

@@ -0,0 +1,100 @@
+import type { MemoryEntry } from './entries.js';
+import { scanProjectMemory } from './scan.js';
+import { Bm25Ranker } from './bm25.js';
+import { embed, isEmbeddingAvailable } from './embeddings.js';
+
+const SEARCH_MODE = process.env['MEMORY_SEARCH'] ?? 'hybrid';
+
+function extractKeywords(query: string): string[] {
+  return query
+    .toLowerCase()
+    .replace(/[^a-z0-9\s]/g, '')
+    .split(/\s+/)
+    .filter((w) => w.length > 2);
+}
+
+export function rankByRelevance(query: string, entries: MemoryEntry[]): MemoryEntry[] {
+  const keywords = extractKeywords(query);
+  if (keywords.length === 0) return entries.slice(0, 5);
+
+  const scored = entries.map((entry) => {
+    let score = 0;
+    const searchText = `${entry.title} ${entry.content} ${entry.tags.join(' ')}`.toLowerCase();
+    for (const kw of keywords) {
+      if (entry.title.toLowerCase().includes(kw)) score += 3;
+      if (entry.tags.some((t) => t.toLowerCase().includes(kw))) score += 2;
+      if (entry.content.toLowerCase().includes(kw)) score += 1;
+    }
+    return { entry, score };
+  });
+
+  return scored
+    .filter((s) => s.score > 0)
+    .sort((a, b) => b.score - a.score)
+    .slice(0, 10)
+    .map((s) => s.entry);
+}
+
+export async function rankByHybrid(
+  query: string,
+  entries: MemoryEntry[],
+): Promise<MemoryEntry[]> {
+  if (entries.length === 0) return [];
+  const texts = entries.map((e) => `${e.title} ${e.content} ${e.tags.join(' ')}`);
+
+  const bm25 = new Bm25Ranker();
+  bm25.fit(texts);
+  const bm25Scores = texts.map((_, i) => bm25.score(query, i));
+  const maxBm25 = Math.max(...bm25Scores, 1);
+  const normBm25 = bm25Scores.map((s) => s / maxBm25);
+
+  let cosineScores: number[] = [];
+  if (isEmbeddingAvailable()) {
+    const vectors = await embed([query, ...texts]);
+    if (vectors) {
+      const queryVec = vectors[0]!;
+      cosineScores = texts.map((_, i) => {
+        const vec = vectors[i + 1];
+        if (!vec) return 0;
+        let dot = 0, nA = 0, nB = 0;
+        for (let j = 0; j < queryVec.length; j++) {
+          dot += queryVec[j]! * vec[j]!;
+          nA += queryVec[j]! * queryVec[j]!;
+          nB += vec[j]! * vec[j]!;
+        }
+        const denom = Math.sqrt(nA) * Math.sqrt(nB);
+        return denom === 0 ? 0 : dot / denom;
+      });
+    }
+  }
+
+  const scored = entries.map((entry, i) => {
+    const combined = (normBm25[i]! * 0.3) + ((cosineScores[i] ?? 0) * 0.7);
+    return { entry, score: combined };
+  });
+
+  return scored
+    .filter((s) => s.score >= 0.15)
+    .sort((a, b) => b.score - a.score)
+    .slice(0, 10)
+    .map((s) => s.entry);
+}
+
+export async function loadMemoryForSession(
+  projectRoot: string,
+  _sessionId?: string,
+  query?: string,
+): Promise<string[]> {
+  const entries = await scanProjectMemory(projectRoot);
+  if (entries.length === 0) return [];
+
+  const relevant = query
+    ? SEARCH_MODE === 'keyword'
+      ? rankByRelevance(query, entries)
+      : await rankByHybrid(query, entries)
+    : entries.slice(0, 5);
+
+  return relevant.map((e) => `[${e.topic}] ${e.title}: ${e.content}`);
+}
+
+export { initEmbeddings } from './embeddings.js';

apps/server/src/services/memory/scan.ts

@@ -0,0 +1,72 @@
+import { homedir } from 'node:os';
+import { join } from 'node:path';
+import { readFile, readdir } from 'node:fs/promises';
+import type { MemoryEntry } from './entries.js';
+import { parseMemoryEntries } from './entries.js';
+import { getMemoryRoot } from './paths.js';
+
+export interface MemoryScope {
+  projectRoot: string;
+  sessionDir?: string;
+  homeDir?: string;
+}
+
+async function scanDirectory(dir: string): Promise<MemoryEntry[]> {
+  const entries: MemoryEntry[] = [];
+  try {
+    const files = await readdir(dir, { withFileTypes: true });
+    for (const file of files) {
+      if (file.isFile() && file.name.endsWith('.md')) {
+        const content = await readFile(join(dir, file.name), 'utf8');
+        entries.push(...parseMemoryEntries(file.name, content));
+      }
+    }
+  } catch {
+    // Directory doesn't exist
+  }
+  return entries;
+}
+
+const MEMORY_TOPICS = ['project', 'user', 'reference'] as const;
+
+async function scanTopicDirs(root: string): Promise<MemoryEntry[]> {
+  const entries: MemoryEntry[] = [];
+  for (const topic of MEMORY_TOPICS) {
+    entries.push(...(await scanDirectory(join(root, topic))));
+  }
+  return entries;
+}
+
+export async function scanMemoryScopes(scope: MemoryScope): Promise<MemoryEntry[]> {
+  const allEntries: MemoryEntry[] = [];
+
+  // 1. Global (~/.boocode/memory/) - lowest priority
+  allEntries.push(...(await scanTopicDirs(getMemoryRoot(homedir()))));
+
+  // 2. Home ($HOME/.boocode/memory)
+  const homeDir = scope.homeDir ?? homedir();
+  const homeRoot = getMemoryRoot(homeDir);
+  if (homeRoot !== getMemoryRoot(homedir())) {
+    allEntries.push(...(await scanTopicDirs(homeRoot)));
+  }
+
+  // 3. Project (.boocode/memory/ under project root)
+  allEntries.push(...(await scanTopicDirs(getMemoryRoot(scope.projectRoot))));
+
+  // 4. Session (.boocode/sessions/<id>/memory.md) - highest priority
+  if (scope.sessionDir) {
+    try {
+      const sessionFile = join(scope.sessionDir, 'memory.md');
+      const content = await readFile(sessionFile, 'utf8');
+      allEntries.push(...parseMemoryEntries('session-memory', content));
+    } catch {
+      // No session memory file
+    }
+  }
+
+  return allEntries;
+}
+
+export async function scanProjectMemory(projectRoot: string): Promise<MemoryEntry[]> {
+  return scanMemoryScopes({ projectRoot });
+}

apps/server/src/services/memory/store.ts

@@ -0,0 +1,35 @@
+import { readFile, writeFile, readdir } from 'node:fs/promises';
+import { join } from 'node:path';
+import type { MemoryTopic } from './paths.js';
+import { getTopicDir } from './paths.js';
+
+export async function readTopicFiles(root: string, topic: MemoryTopic): Promise<Map<string, string>> {
+  const dir = getTopicDir(root, topic);
+  const files = new Map<string, string>();
+  try {
+    const entries = await readdir(dir, { withFileTypes: true });
+    for (const entry of entries) {
+      if (entry.isFile() && entry.name.endsWith('.md')) {
+        const content = await readFile(join(dir, entry.name), 'utf8');
+        files.set(entry.name, content);
+      }
+    }
+  } catch {
+    // Directory doesn't exist yet
+  }
+  return files;
+}
+
+export async function writeEntry(
+  root: string,
+  topic: MemoryTopic,
+  title: string,
+  content: string,
+  tags: string[],
+): Promise<void> {
+  const dir = getTopicDir(root, topic);
+  const tagLine = tags.length > 0 ? `> tags: ${tags.join(', ')}\n\n` : '\n';
+  const entry = `## ${topic}: ${title}\n${tagLine}${content}\n`;
+  const filename = title.toLowerCase().replace(/[^a-z0-9]+/g, '-').replace(/(^-|-$)/g, '') + '.md';
+  await writeFile(join(dir, filename), entry, 'utf8');
+}

apps/server/src/services/synthesisPipeline.ts

@@ -35,6 +35,7 @@ export const SYNTHESIS_TOOLS: ReadonlySet<string> = new Set([
   'get_codebase_overview',
   'get_framework_analysis',
   'get_semantic_neighborhoods',
+  'get_blast_radius',
 ]);
 
 const TOP_N_FILES = 5;

apps/server/src/services/system-prompt.ts

@@ -22,6 +22,8 @@ import { readFile, stat } from 'node:fs/promises';
 import type { Agent, Project, Session } from '../types/api.js';
 import { getAgentsMtimes } from './agents.js';
 import { resolveRoute } from './inference/provider.js';
+import { loadMemoryForSession } from './memory/recall.js';
+import { formatMemoryBlock } from './memory/prompt.js';
 
 const BASE_SYSTEM_PROMPT = (projectPath: string) =>
   `You are BooCode Chat, a code investigation assistant. The user is working on a project located at ${projectPath}. Use the file-read tools (view_file, list_dir, grep, find_files) to investigate code when needed. Be concise. Cite file paths and line numbers when discussing code. Do not hallucinate file contents — read the file first. Tool results may be truncated; if so, narrow your query rather than guessing.`;
@@ -164,7 +166,11 @@ export async function buildSystemPromptWithFingerprint(
   let out = BASE_SYSTEM_PROMPT(project.path);
   const guidance = await getContainerGuidance();
   if (guidance) {
-    out += `\n\n--- Container guidance ---\n${guidance}\n--- end container guidance ---\n`;
+    out += '\n\n--- Container guidance ---\n' + guidance + '\n--- end container guidance ---\n';
+  }
+  const memory = await loadMemoryForSession(project.path, session.id).catch(() => []);
+  if (memory.length > 0) {
+    out += '\n\n' + formatMemoryBlock(memory);
   }
   if (agent && agent.system_prompt.trim().length > 0) {
     out += '\n\n' + agent.system_prompt.trim();

apps/server/src/services/tools/codecontext/get_call_graph.ts

@@ -0,0 +1,31 @@
+import { z } from 'zod';
+import { makeCodecontextTool } from './factory.js';
+
+export const GetCallGraphInput = z.object({
+  symbol: z.string().describe('Symbol name to analyze'),
+  depth: z.number().int().min(1).max(5).optional().describe('Max traversal depth (default 2)'),
+});
+export type GetCallGraphInputT = z.infer<typeof GetCallGraphInput>;
+
+const DESCRIPTION =
+  'Returns a call graph for a function or method: callers, callees, and transitive references. ' +
+  'Use to understand how a symbol is invoked and what it depends on.';
+
+const { toolDef: getCallGraph, execute: executeGetCallGraph } =
+  makeCodecontextTool<GetCallGraphInputT>({
+    name: 'get_call_graph',
+    schema: GetCallGraphInput,
+    description: DESCRIPTION,
+    jsonParameters: {
+      type: 'object',
+      properties: {
+        symbol: { type: 'string', description: 'Symbol name to analyze' },
+        depth: { type: 'number', description: 'Max traversal depth (default 2)' },
+      },
+      required: ['symbol'],
+      additionalProperties: false,
+    },
+    mapArgs: (input) => ({ symbol: input.symbol, depth: input.depth ?? 2 }),
+  });
+
+export { getCallGraph, executeGetCallGraph };

apps/server/src/services/tools/codecontext/get_symbol_details.ts

@@ -0,0 +1,31 @@
+import { z } from 'zod';
+import { makeCodecontextTool } from './factory.js';
+
+export const GetSymbolDetailsInput = z.object({
+  symbol: z.string().describe('Symbol name to resolve'),
+  file_path: z.string().optional().describe('Optional file path to narrow search'),
+});
+export type GetSymbolDetailsInputT = z.infer<typeof GetSymbolDetailsInput>;
+
+const DESCRIPTION =
+  'Returns type signature, definition location, and usage count for a named symbol. ' +
+  'Use after get_codebase_overview to dive deeper into specific functions, classes, or variables.';
+
+const { toolDef: getSymbolDetails, execute: executeGetSymbolDetails } =
+  makeCodecontextTool<GetSymbolDetailsInputT>({
+    name: 'get_symbol_details',
+    schema: GetSymbolDetailsInput,
+    description: DESCRIPTION,
+    jsonParameters: {
+      type: 'object',
+      properties: {
+        symbol: { type: 'string', description: 'Symbol name to resolve' },
+        file_path: { type: 'string', description: 'Optional file path to narrow search' },
+      },
+      required: ['symbol'],
+      additionalProperties: false,
+    },
+    mapArgs: (input) => ({ symbol: input.symbol, file_path: input.file_path }),
+  });
+
+export { getSymbolDetails, executeGetSymbolDetails };

apps/server/src/services/tools/extract_memory.ts

@@ -0,0 +1,44 @@
+import { z } from 'zod';
+import type { ToolDef } from '../tools/types.js';
+import { ensureMemoryScaffold, getMemoryRoot } from '../memory/paths.js';
+import { writeEntry } from '../memory/store.js';
+
+const ExtractMemoryInput = z.object({
+  topic: z.enum(['project', 'user', 'reference']).describe('Memory topic category'),
+  title: z.string().min(1).max(200).describe('Entry title (will be normalized to filename)'),
+  content: z.string().min(1).describe('Memory content body'),
+  tags: z.array(z.string()).optional().describe('Optional tags for search'),
+});
+
+type InputT = z.infer<typeof ExtractMemoryInput>;
+
+export const extractMemoryTool: ToolDef<InputT> = {
+  name: 'extract_memory',
+  description: 'Persist a memory entry to .boocode/memory/ for cross-session recall. Use for project conventions, user preferences, and architectural decisions.',
+  inputSchema: ExtractMemoryInput,
+  jsonSchema: {
+    type: 'function',
+    function: {
+      name: 'extract_memory',
+      description: 'Persist a memory entry for cross-session recall',
+      parameters: {
+        type: 'object',
+        properties: {
+          topic: { type: 'string', enum: ['project', 'user', 'reference'] },
+          title: { type: 'string', description: 'Entry title' },
+          content: { type: 'string', description: 'Memory content' },
+          tags: { type: 'array', items: { type: 'string' }, description: 'Search tags' },
+        },
+        required: ['topic', 'title', 'content'],
+      },
+    },
+  },
+  async execute(input: InputT, projectRoot: string): Promise<unknown> {
+    const root = getMemoryRoot(projectRoot);
+    await ensureMemoryScaffold(root);
+    await writeEntry(root, input.topic, input.title, input.content, input.tags ?? []);
+    return {
+      result: `Memory entry "${input.title}" saved to .boocode/memory/${input.topic}/`,
+    };
+  },
+};

apps/server/src/services/tools/search_memory.ts

@@ -0,0 +1,40 @@
+import { z } from 'zod';
+import type { ToolDef } from '../tools/types.js';
+import { scanProjectMemory } from '../memory/scan.js';
+import { rankByHybrid } from '../memory/recall.js';
+
+const SearchMemoryInput = z.object({
+  query: z.string().min(1).describe('Search query to match against memory entries'),
+});
+
+type InputT = z.infer<typeof SearchMemoryInput>;
+
+export const searchMemoryTool: ToolDef<InputT> = {
+  name: 'search_memory',
+  description: 'Search the .boocode/memory/ store for relevant entries. Returns ranked results matching the query. Use before asking about project conventions or preferences.',
+  inputSchema: SearchMemoryInput,
+  jsonSchema: {
+    type: 'function',
+    function: {
+      name: 'search_memory',
+      description: 'Search memory store for relevant entries',
+      parameters: {
+        type: 'object',
+        properties: {
+          query: { type: 'string', description: 'Search query' },
+        },
+        required: ['query'],
+      },
+    },
+  },
+  async execute(input: InputT, projectRoot: string): Promise<unknown> {
+    const entries = await scanProjectMemory(projectRoot);
+    if (entries.length === 0) return { result: 'No memory entries found.' };
+
+    const relevant = await rankByHybrid(input.query, entries);
+    if (relevant.length === 0) return { result: 'No matching memory entries.' };
+
+    const lines = relevant.map((e) => `[${e.topic}] ${e.title}: ${e.content}`);
+    return { result: `Found ${relevant.length} entry(ies):\n${lines.join('\n')}` };
+  },
+};

apps/web/src/App.tsx

@@ -16,6 +16,7 @@ import { RightRailDrawerProvider, useRightRailDrawer } from '@/hooks/useRightRai
 import { useViewport } from '@/hooks/useViewport';
 import { ThemeFx } from '@/components/fx/ThemeFx';
 import { FlowLauncherDialog } from '@/components/FlowLauncherDialog';
+import { ArenaLauncherDialog } from '@/components/ArenaLauncherDialog';
 
 function SessionRightRail() {
   const { id } = useParams<{ id: string }>();
@@ -102,6 +103,7 @@ function AppShell() {
         </Routes>
         <Toaster position="bottom-right" />
         <FlowLauncherDialog />
+        <ArenaLauncherDialog />
       </div>
     </>
   );