The Agent Skills Directory

[COMMAND_EXECUTION]: Multiple Python scripts within the skill utilize the subprocess module to perform system tasks. scripts/run_eval.py and scripts/improve_description.py execute the claude CLI to run subagent evaluations and improve skill descriptions. Additionally, eval-viewer/generate_review.py runs the lsof utility to identify and manage occupied network ports.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes user-defined skill drafts and test prompts without automated sanitization, though it mitigates this with clear security instructions.
Ingestion points: Processes user-provided intents and skill drafts in SKILL.md and test queries in evals/evals.json.
Boundary markers: Uses the 'eval-viewer' to facilitate manual human review of all generated content.
Capability inventory: Utilizes subprocess for executing the claude CLI and has broad file system read/write access.
Sanitization: Lacks explicit input sanitization, relying instead on integrated 'Garde-fous de sécurité' guidelines that warn against creating destructive or malicious skills.

skill-creator