The Agent Skills Directory

[SAFE]: The skill was analyzed for common attack vectors, including prompt injection and data exfiltration. No malicious patterns were identified. The code exhibits strong defensive programming, such as input validation for repository names and human-in-the-loop gates for all remote operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from external GitHub repositories to perform its persona analysis. It implements mitigation strategies by instructing the AI agent to treat target repository content strictly as data and providing functions in persona.ts to sanitize input by stripping metadata, such as HTML comments, that could contain hidden instructions.
[COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh) via Node.js execFileSync. The repository name inputs are validated against a restrictive regular expression in measure.ts to prevent command injection, ensuring shell execution is limited to legitimate repository measurement.
[DATA_EXFILTRATION]: A security gate script, check-draft.ts, is included to protect against accidental metadata leakage. It identifies if a submission for a public repository contains references to private repositories or internal hostnames, preventing sensitive information from being published.

maintainer-persona