The Agent Skills Directory

[SAFE]: The skill's instructions and playbooks demonstrate a strong security posture by explicitly distinguishing between documentation (guidance) and enforcement (deterministic gates). It repeatedly warns against the 'docs-as-reliability-fix' trap, directing users to move safety-critical rules to CI gates or hooks rather than relying on agent adherence to markdown instructions.
[PROMPT_INJECTION]: The skill has a documented attack surface for indirect prompt injection because it is designed to ingest and analyze external repository files like AGENTS.md and llms.txt. However, this risk is mitigated by the skill's specific focus on auditing and reporting rather than execution.
Ingestion points: The skill reads untrusted documentation files from the target repository, including AGENTS.md, llms.txt, and tool descriptions.
Boundary markers: The playbooks (e.g., in references/playbooks/agents-md.md) emphasize that documentation is a contract for guidance and that mandatory invariants must be backed by deterministic gates (CI/hooks) handled by separate hardening skills.
Capability inventory: The skill is limited to reading files and writing audit results, plans, and ledgers to specific directory paths (docs/audits/ or audit-artifacts/). It does not possess capabilities for arbitrary system commands or unauthorized network transmission.
Sanitization: The prompts focus on structural and clarity audits rather than literal interpretation of the content as commands.

agent-docs