The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Python scripts to manage its optimization and evaluation cycles. This involves executing scripts like execute.py and eval_runner.py within the local environment to automate skill refinements.
[PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection because its primary purpose is to ingest, evaluate, and modify instructions from other skill files.
Ingestion points: Files targetted by the --target parameter in the execute.py script, such as other skills' SKILL.md files.
Boundary markers: None identified in the skill instructions; the agent is directed to process the content of target skills directly.
Capability inventory: The skill is granted Bash, Read, and Write capabilities, and it executes Python scripts that can modify repository content.
Sanitization: No specific sanitization or filtering logic is mentioned for the external skill data being processed.

exploration-optimizer