The Agent Skills Directory

[COMMAND_EXECUTION]: Executes local shell commands including git ls-files, git log, and grep to perform mechanical scans of the repository for stale or inconsistent documentation.
[COMMAND_EXECUTION]: Invokes the GitHub CLI (gh) to check the status of project issues and pull requests, assisting in identifying completed but un-archived documentation plans.
[DATA_EXFILTRATION]: Performs network requests to GitHub's official API to verify issue and pull request states. These operations target a well-known service and do not transmit sensitive information such as local environment secrets.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from documentation and source code files, which are potentially untrusted inputs that could contain hidden instructions.
Ingestion points: Reads and audits markdown files in the docs/ directory and referenced source code modules.
Boundary markers: Uses <doc> tags in prompt templates to separate document content from agent instructions.
Capability inventory: Includes the ability to propose and execute file updates, deletions, and migrations.
Sanitization: The skill enforces a mandatory human approval gate in Phase 3 before any file modifications or deletions are executed, effectively mitigating risks associated with untrusted data processing.

docs-freshness-audit