The Agent Skills Directory

[SAFE]: Boundary Governance: The skill establishes non-negotiable protection for sensitive directories like .memory/ and .agents/, ensuring operational state and agent infrastructure remain unmodified without explicit human permission.\n- [INDIRECT_PROMPT_INJECTION]: Trusted Data Handling: While the skill ingests untrusted repository content, it provides clear instructions to treat this data as passive content and ignore any embedded directives, mitigating the risk of indirect prompt injection.\n- [PROMPT_INJECTION]: Guideline Verification: The automated detection of prompt injection is a false positive; the instructions in references/SECURITY.md are defensive in nature and guide the agent to resist injection attempts.\n- [SAFE]: Secret Isolation: The Zero Secret Exposure Protocol mandates that any tokens or keys detected during the synchronization process must be redacted, ensuring no sensitive information is committed to the repository or logs.\n- [SAFE]: Verification Lifecycle: Every modification is subjected to a 14-point audit checklist and a bulk-edit verification gate (Step 21), which includes diff audits and mechanical checks to maintain codebase integrity.

updatedocs