The Agent Skills Directory

[COMMAND_EXECUTION]: Several scripts (scripts/platforms/claude_code.py, scripts/platforms/codex.py, scripts/platforms/openclaw.py) use subprocess.run() or subprocess.Popen() to interact with local agent CLIs. These are intended for the primary purpose of triggering and evaluating skills on different platforms.
[METADATA_POISONING]: The SKILL.md description is intentionally 'pushy' (as noted in the skill's own instructions) to ensure the AI agent triggers correctly. While this is a form of instruction manipulation, it is the explicitly stated goal of this developer-centric skill and serves a functional purpose for testing trigger accuracy.
[INDIRECT_PROMPT_INJECTION]: The skill processes eval queries and user feedback to refine skill descriptions. While this is an ingestion point for untrusted data, the skill is designed to be used in a development environment where the user is refining their own content, and the capabilities (script generation, benchmarking) are within the expected scope of a 'skill creator' tool.

skill-creator