The Agent Skills Directory

[SAFE]: The instructions describe a legitimate, multi-step document review process with clear boundaries and user approval requirements for significant modifications.
[DATA_EXPOSURE]: The skill includes a 'Security' lens designed to detect and flag potential data exposure risks (PII, tokens, credentials) in the documents being reviewed. This is a defensive feature intended to help the user identify security gaps in their own documentation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document content which presents a standard ingestion attack surface.
Ingestion points: SKILL.md Step 1 reads user-provided documents or files from the docs/ directory.
Boundary markers: Not explicitly defined; document content is processed as the primary input.
Capability inventory: The skill has read/write access to local documents (Step 1, Step 6) and the ability to dispatch sub-agents for reading tests (Step 7).
Sanitization: No explicit sanitization or escaping of document content is mentioned before processing.
[COMMAND_EXECUTION]: SPEC.md contains example shell commands for local evaluation and testing (e.g., python3 distillery/scripts/distiller.py). These are developer-facing instructions for maintenance and do not represent autonomous or malicious runtime execution by the agent.

ia-document-review