The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.Popen and subprocess.run to execute local Python scripts and the official claude CLI. These operations are used for aggregating benchmark data, packaging skills into zip archives, and running evaluation queries against a local project environment.
[EXTERNAL_DOWNLOADS]: The scripts/improve_description.py script utilizes the anthropic Python library to communicate with the Anthropic API. This is used for iterative refinement of skill descriptions based on test results.
[DATA_EXPOSURE]: The eval-viewer/generate_review.py script launches a local HTTP server on 127.0.0.1:3117. This server provides a web interface for users to review the outputs of skill evaluations and is restricted to localhost access.
[INDIRECT_PROMPT_INJECTION]: As a developer tool that processes user-provided skill drafts and test queries, the skill has an inherent surface for indirect prompt injection. However, it employs boundary markers (e.g., XML-like tags) in its optimization prompts to mitigate accidental instruction obedience from untrusted data.

skill-creator