The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates agents to write and run diagnostic tests and code patches. Specifically, the 'Warden' and 'Artificer' roles execute tests and scripts within the local development environment to reproduce bugs and verify the effectiveness of the fixes.
[DATA_EXFILTRATION]: To analyze defects, the skill reads local project metadata, dependency manifests, and error logs. It incorporates a non-negotiable security rule that forbids the inclusion of secrets, tokens, or PII in agent context, messages, or checkpoint files.
[PROMPT_INJECTION]: The skill processes external data such as user bug descriptions and log files. While this creates a surface for indirect prompt injection, the risk is mitigated by a multi-phase 'skeptic gate' architecture that requires a separate Opus-level model to challenge and approve all deliverables before the agent can advance to the next phase.

squash-bugs