The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill manages its own environment by installing well-known, standard testing packages pytest and pytest-xdist using uv pip or pip if they are not already present. This is a common practice for development-oriented tools.
[COMMAND_EXECUTION]: The skill executes local shell commands to retrieve MLflow traces and run the pytest suite. These commands are essential to the primary function of diagnosing and verifying agent behavior fixes.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions, representing a vulnerability surface.
Ingestion points: The skill reads agent execution traces (mlflow traces get) and incorporates raw user feedback into trace metadata (mlflow.log_feedback).
Boundary markers: The instructions direct the agent to quote and restate observations, but do not specify technical delimiters to separate untrusted trace content from the analyzer's instructions.
Capability inventory: The skill can write files to the local system, execute Python code via pytest, and communicate with MLflow tracking servers.
Sanitization: No specific sanitization or filtering of the ingested trace content or user rationale is mentioned.
[DYNAMIC_EXECUTION]: The skill generates Python test scripts based on user requirements and executes them at runtime to verify agent behavior. This dynamic generation and execution of code is the intended mechanism for the skill's regression testing workflow.

fix-agent-issue