The Agent Skills Directory

[SAFE]: No malicious instructions, prompt injections, or attempts to bypass safety filters were detected in the instructions or templates.
[SAFE]: The skill does not perform unauthorized network operations, file exfiltration, or access sensitive system credentials. It operates on project-specific files like plan.json and spec.md as expected for a coding tool.
[SAFE]: No remote code execution patterns, external package installations, or obfuscated content were found.
[SAFE]: The subagent interaction pattern creates a processing surface for external task data, but the skill includes robust defensive prompting (e.g., instructing reviewers to independently verify code changes rather than trusting summaries), which effectively mitigates risks associated with processing untrusted agent outputs.

code-subagents