The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to autonomously execute various build, lint, and test commands including npm, go, pytest, cargo, mvn, and make based on project stack detection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted console output from failed verification steps (lint/build/test errors) to determine how to modify the source code. A malicious test file or script could produce output designed to influence the agent's behavior during the autonomous fixing phase. * Ingestion points: Standard output and error streams from verification commands in the self-healing loop. * Boundary markers: None provided to distinguish error data from agent instructions. * Capability inventory: File writing (to apply fixes) and command execution (to retry the loop). * Sanitization: None present.
[DYNAMIC_EXECUTION]: The skill generates and applies source code modifications at runtime to resolve errors and then immediately executes the altered code, creating a feedback loop where the agent runs code it has dynamically generated.

gate