メニュー
Search the project's markdown knowledge base via the GitMark CLI instead of grep cmd. Use to find where something is documented, answer "where do the docs say X", or before reading many files at random; Hits come back as `file:line · heading · snippet`.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Feature/fix development procedure for this framework: proposal → research → technical concept → tasks (/goal) → worktree → implementation (human-in-the-loop) → pre-commit/CI/PR → review → deploy. The skill is about HOW the agent interacts with the user at each stage. Use it when a feature/fix starts or someone asks "how do we develop here".
Rules for maintaining a markdown knowledge base (KB/GitMark) — apply when adding, editing, moving, or deleting documentation (.md). Keeps the KB structured instead of a pile of files. Use on "add a doc", "record a decision", "update the docs", "reorganize docs", "rebuild the KB"
Slice the accumulated feature context (proposal + research + HLD + the fixed stack) into a task list for the agent: self-contained units written "by an agent for agents", with a mandatory DoD and list of checks, explicit dependencies, ready for /goal. Use at the dev-flow task-creation stage — after design, before implementation.
Roll out the whole very-ai-framework (dev-flow process + GitMark knowledge base + CI/CD) from the template repo into a target project: copy skills/commands into .claude/, scaffold the KB, port the GitHub Actions pipeline, and configure environments / secrets / deploy. Use to set up the framework in a new or different repository.
| name | kb-search |
| description | Search the project's markdown knowledge base via the GitMark CLI instead of grep cmd. Use to find where something is documented, answer "where do the docs say X", or before reading many files at random; Hits come back as `file:line · heading · snippet`. |
The repo's markdown is a md + README-index + git knowledge base. Markdown is the source of truth; the search index is derived and regenerated from it
G="python3 .claude/skills/kb-search/gitmark.py" # paths are relative to the project root
/kb-search <query>)$G index (rebuild, fast).$G search "<query>" -k 8 — [bm25] exact term · [trigram] substring · [fuzzy] typos/forms/Cyrillic.file:line), open the 1–2 most relevant, and answer — don't reprint whole snippets./kb-search (no query) → $G stat + show the /kb-search <query> syntax.--json for machine-readable output.plan/report: stale plans, one-off audits).
Add --scope all to include them, or --scope history for only those — use it when the
question is about past intent or a one-off analysis rather than current state (e.g. "what
was the original plan for X", "what was the context when we scoped X", "what did the audit
of Z find")./kb-graph (separate script graph.py; HTML ontology graph)..md, never the index..gitmark/ — it's a gitignored, rebuildable cache.kb-maintain skill.