The Agent Skills Directory

[SAFE]: The skill provides a structured methodology for auditing system architecture, state management, and concurrency models. The instructions focus on documentation and analysis rather than execution of untrusted code or exfiltration of sensitive information.
[INDIRECT_PROMPT_INJECTION]: The skill creates a potential surface for indirect prompt injection because it is designed to process external documentation, issue trackers, and user-provided references. However, this risk is thoroughly mitigated by the comprehensive guidelines in references/validating-claims.md, which instruct the agent to treat external claims as leads rather than facts and to prioritize primary evidence (such as source code and logs) to confirm any reported defects before including them in the research artifacts.

antithesis-research