v0.10.13 — works with codegraph

Open Second Brain v0.10.13 — works with codegraph

Open Second Brain now works alongside the standalone codegraph CLI when it is present on the host. The agent picks codegraph_* tools for structural code questions (callers, callees, impact, "where is X"), falls back to a grep+read loop with an argued install recommendation when codegraph is absent, and surfaces the same state through o2b doctor so operators see it too. CheckResult contract unchanged; codegraph keeps its own lifecycle, storage, and MCP registration — Open Second Brain only detects and routes.

Process wins

• Token economy: structural questions stop spawning grep+read loops.
• Accuracy: AST-precise hits replace text-greedy grep false positives.
• Tool discipline: each question shape maps to one tool, no double-checking with grep.
• Operator visibility: o2b doctor reports codegraph health alongside the rest.
• Safety net: hard rule appends .codegraph/ to .gitignore after init, so the local index never lands in a commit.
• Argued recommendation, not nag: one paragraph with the real file count from the user's project, dropped if the user says no.

What ships

• skills/codegraph-partner/SKILL.md — agent-facing playbook (detect → branch → act + disambiguation table + .gitignore rule after codegraph init).
• code_graph check in o2b doctor (cwd + vault parent siblings, capped at 50 dirs, configurable via DoctorOptions.partner.codegraph).
• src/core/partner/codegraph.ts — detection module with DI-friendly hooks.

Notes

• Open Second Brain never calls codegraph install / init / index. The CLI is detected, never driven.
• .codegraph/ is now in .gitignore.