Maintains project .codex/rules and the AGENTS.md rule index. Explicit user invocation only.
81%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Use only when the user explicitly requests rule creation, synchronization, improvement, review, or inventory.
.codex/rules/; never use personal or foreign-assistant rule paths.AGENTS.md files as Codex's automatically loaded instruction surface.AGENTS.md contains a compact index, not copied rule bodies.Position 1 of the arguments is a free-form prompt (RU/EN); mode tokens are optional and may follow. Resolve the mode from that prompt: status (default, read-only inventory), list, create, improve, review. An explicit mode token wins outright; otherwise take the mode the prompt describes, and ask at most one clarifying question, only when the answer changes what gets written.
Before the first write — and before the report on a read-only run — state in one block: the input verbatim, the resolved mode and why, the scope (rule files and AGENTS.md tables in play), what will be done, and the result the user ends up with.
AGENTS.md files and inventory every project rule with rg --files .codex/rules | sort.status, create, improve, review, list, or synchronize. Persist durable repository facts only; exclude secrets, transient session state, generated reports, and duplicate instructions..codex/rules/. Reuse an existing rule when its scope matches.AGENTS.md rule-index table. The repository-root table lists every .codex/rules/*.md file exactly once with columns Rule, Load when, and Purpose. Nested tables add only subtree-specific rules not already covered by the inherited root index.$brewtools:text-optimize -l for every changed rule and AGENTS.md index. Preserve paths, scope qualifiers, safety constraints, and table structure; apply the optimized patch only after comparing semantics.AGENTS.md.The index makes rules discoverable; it does not auto-load their bodies. During later work, read the indexed rule whose Load when condition matches the task.
Required approval: main presents a concrete, reviewable proposal in chat and waits for an actual user reply before dependent action. Existing authorization for the same scope remains valid; do not ask again. Optional clarification: use request_user_input_async only if exposed, or request_user_input only if available in the current runtime/mode, for optional choices and never approval. Otherwise ask in main chat. Delegated agents return unresolved questions to main. Silence, elapsed time and tool errors are not approval.
kochetkov-ma/claude-brewcode
5da825f
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.