The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to review user issues and reports to understand reader problems. These external data sources represent an indirect prompt injection surface as they could contain malicious instructions. The skill mitigates this by requiring all claims to be verified against the repository's source code, tests, and CLI help, and explicitly prohibits inventing documentation based solely on external threads.
Ingestion points: "Issues and user reports" mentioned in the Sources of Truth section.
Boundary markers: Absent.
Capability inventory: Modifies local documentation files (README.md, MDX) via file system writes.
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill directs the agent to execute pnpm --filter @repo/docu checktypes to verify documentation before shipping. This is a standard, task-appropriate development command used to ensure type safety in documentation frameworks and does not involve arbitrary or malicious shell execution.

technical-writing