The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to verify repository state, including git operations to check branches and commit status. It runs a repository-local script located at .claude/scripts/review-state.js or scripts/review-state.js using Node.js to derive compliance metrics.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it is designed to ingest and repeat project rules to the agent. 1. Ingestion points: CLAUDE.md and project-specific rule files in the rules/ directory. 2. Boundary markers: Content is quoted within markdown blockquotes. 3. Capability inventory: The skill can execute shell commands via Bash and trigger other automated workflows via the Skill tool. 4. Sanitization: Rule content is quoted directly into the prompt without sanitization or escaping.
[COMMAND_EXECUTION]: The skill demonstrates high autonomy by design, dynamically invoking other tools like /codex-review-fast based on the results of its state analysis without requiring additional user confirmation.

remind