The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill executes standard, read-only commands such as ls, cat, jq, and git status to verify the state of the local documentation directory. No destructive or high-privilege commands are present.
[EXTERNAL_DOWNLOADS] (SAFE): The provided GitHub Action workflow template references official actions from the 'actions' organization (e.g., actions/checkout@v4). These are trusted sources according to the [TRUST-SCOPE-RULE].
[INDIRECT_PROMPT_INJECTION] (LOW): The skill reads content from local markdown and JSON files (plugins/compound-engineering/agents/*.md, etc.).
Ingestion points: Multiple documentation files and JSON manifests.
Boundary markers: None present.
Capability inventory: File system read, JSON parsing, and git status reporting.
Sanitization: None detected.
Assessment: While it processes untrusted local data, the lack of dangerous capabilities (network write, execution of data as code) makes the risk negligible.

deploy-docs