The Agent Skills Directory

[SAFE]: No malicious activity was detected. The skill is an auditing tool that operates strictly on local files and configuration within the user's project directory.
[COMMAND_EXECUTION]: The coordinator uses standard system commands such as 'date' for run ID generation and performs local filesystem operations (directory creation, file reading, and writing) necessary for its scanning and state management functions as defined in SKILL.md and STATE.md.
[PROMPT_INJECTION]: The skill audits untrusted codebase data, creating a surface for indirect prompt injection. 1. Ingestion points: Source and configuration files in the target repository (SKILL.md). 2. Boundary markers: Sub-agents are directed to specific file paths via structured input manifests (FORMAT.md). 3. Capability inventory: Sub-agents (Haiku) can search for patterns and return excerpts; the coordinator reads/writes files and verifies evidence (STATE.md). 4. Sanitization: The 'Iron Law' verification gate requires the coordinator to re-verify all excerpts against the original files, which prevents sub-agents from fabricating proof or being manipulated into misreporting results (GOLDEN-RULES.md).

prod-readiness