The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates automated testing and optimization by executing the agent's command-line interface (e.g., claude) using subprocess in scripts/improve_description.py and scripts/run_eval.py. This allows it to programmatically evaluate skill descriptions and trigger logic.
[COMMAND_EXECUTION]: The script eval-viewer/generate_review.py uses subprocess.run to invoke the lsof utility for checking port availability when launching the local review server.
[DYNAMIC_EXECUTION]: The scripts/init_skill.py script generates boilerplate Python code and markdown documentation to scaffold new skill projects.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-defined prompts and skill descriptions to perform trigger evaluations and automated improvements. 1. Ingestion points: User-defined queries and expectations within evals/evals.json, as well as prompts processed by scripts/run_eval.py. 2. Boundary markers: The skill uses JSON serialization and YAML block scalars to delimit user content when constructing command files. 3. Capability inventory: Includes file system writes for scaffolding and shell command execution via the agent's CLI. 4. Sanitization: Relies on standard JSON and YAML library serialization for content interpolation.

skill-creator