Removed /opt/boocode/AGENTS.md (per-project override) — the project's
agents now resolve from the global /data/AGENTS.md only. Eliminates the
two-files-must-stay-in-sync footgun that surfaced during B.3
verification.
Fix: agents.ts ALL_TOOL_NAMES was a hardcoded 9-item whitelist that
silently filtered any unknown tool name from agent.tools arrays. This
caused web_search/web_fetch (v1.11.8) and the 8 codecontext tools to be
dropped at parse time. Replaced with ALL_TOOLS.map(t => t.name) for
single source of truth. Pre-existing exposure was dormant since no
builtin agent listed web_search; surfaced by adding codecontext.
New /opt/boocode/codecontext/ directory holding the codecontext sidecar
that BooCode's tool wrappers (track B.2) will talk to. No BooCode-side
changes yet — this commit lands the sidecar standalone.
- Dockerfile: multi-stage golang:1.24-alpine → alpine:3.20. Clones
codecontext at v3.2.1 from github.com/nmakod/codecontext (cgo build for
tree-sitter bindings), builds the shim alongside (CGO_ENABLED=0).
- shim.go: stdlib-only Go HTTP server wrapping codecontext's stdio MCP
child. Newline-delimited JSON framing per the MCP transport spec
(NOT LSP-style Content-Length). 8 POST /v1/* endpoints, one per MCP
tool, plus GET /health. Child supervised via child.Wait() goroutine
that os.Exit's on death so the container's restart: unless-stopped
policy fires (Signal(0) on a zombie returns nil and is not a liveness
check — discovered during kill-restart testing).
- go.mod: no third-party deps; future Go security advisories don't apply.
docker-compose service: joins boocode_net (no host port), mounts
/opt:/opt:ro (BooCode projects live at /opt/<slug>, not exclusively
under /opt/projects), healthcheck on /health.
Verified: build clean, healthcheck reports healthy ~15s after up,
multi-project queries return valid markdown, target_dir swap works on
subtree paths. Kill-restart cycle completes in ~200ms with one failed
health poll observed (no misleading "ok" during the gap). Memory: 24.6
MiB after 5 search_symbols calls, 5.6 MiB after 30 min idle — codecontext
releases the per-call graph between target_dir swaps, so the shim doesn't
hold the indexed state.
