The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to audit the repository state. It uses git diff --name-only to identify modified files and runs a local Node.js script located at scripts/review-state.js (or .claude/scripts/review-state.js) to check the status of pre-commit gates. These commands are restricted to the local environment and are consistent with the skill's purpose of verifying code readiness.- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by ingesting repository data and outputs from other agents. However, the risk is mitigated by deterministic logic and structured analysis workflows.
Ingestion points: Modified filenames and repository state retrieved via git diff; outputs from subprocess skills like /risk-assess and /codex-test-review.
Boundary markers: None explicitly defined for isolating filenames from internal instructions.
Capability inventory: Access to Bash, Agent, and Skill tools for analysis and reporting.
Sanitization: The skill applies deterministic sorting, count-based caps (e.g., top-3 or top-10 modules), and structured scoring models to the ingested data.

pre-pr-audit