The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes scripts/audit_agent.py, a local Python utility for static analysis. It identifies implementation defects in agent codebases, such as unbounded loops or missing tool descriptions, by traversing the filesystem and performing regex-based pattern matching without executing the target files.
[EXTERNAL_DOWNLOADS]: The documentation references the @earendil-works/pi-coding-agent package and pi.dev as part of a worked example for implementing an agent harness. These external references are informative and central to the skill's purpose of providing a concrete implementation reference.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface through its auditing script, which processes untrusted external code files. However, the risk is mitigated as the script performs non-executable static analysis and does not interpolate the content into an LLM prompt context.
[SAFE]: No malicious behavior, obfuscation, or unauthorized data access patterns were detected. The skill's primary focus is on improving the security and reliability of agent systems through systematic auditing and best practices.

agent-harness