The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Python commands through the pixi task runner and generates script files in a local scratch/ directory. These actions are used to programmatically interface with the skore library for report analysis.
[SAFE]: The skill enforces a strict separation of concerns, returning proposed changes as text rather than directly modifying the project's journal files. This allows for human oversight of all generated content.
[SAFE]: Defensive prompting is utilized to prevent the agent from using untrusted training data for API calls, mandating the use of a specialized python-api skill for signature verification.
[SAFE]: The skill ingests data from ML reports via report.diagnosis(). While this constitutes a potential indirect prompt injection surface, the risk is negligible as the output is constrained to generating experiment suggestions which require user promotion to become active.
Ingestion points: Data returned from report.diagnosis() in the reports/ workspace.
Boundary markers: None specified for the ingested report content.
Capability inventory: Python script generation and execution via pixi and scratch/ files.
Sanitization: No explicit sanitization or validation of the report content is mentioned.

iterate-from-skore