The Agent Skills Directory

[SAFE]: The skill defines a controlled multi-agent workflow for autonomous code generation and validation. It incorporates explicit 'Red lines' that prevent the AI from bypassing safety and quality checks, such as instructions to stop if the agent attempts to weaken tests or delete failing checks.
[SAFE]: The provided test file loop.test.js serves as an integrity guardrail. It programmatically verifies that the skill's reference files are free of hidden characters (like U+00A0) and that the subagent configurations maintain proper least-privilege tool scoping (e.g., verifying that the 'checker' agent does not have write permissions).
[SAFE]: Analysis of the execution patterns shows no evidence of remote code execution, external network calls to untrusted domains, or sensitive data exfiltration. File operations are restricted to reading project-local configuration and seed files.
[SAFE]: The 'pi' setup instructions involve standard environment configuration (symlinking extensions) and do not manifest as automated privilege escalation or persistence mechanisms.

loop-go