The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of multiple Python and shell scripts to automate the development lifecycle of other skills.
It instructs the execution of python -m scripts.aggregate_benchmark to compile evaluation metrics.
It uses python eval-viewer/generate_review.py to launch a review server or generate static HTML reports for human evaluation.
It invokes python -m scripts.run_loop to perform description optimization, which involves calling the Claude CLI.
It uses python -m scripts.package_skill to bundle the completed skill for distribution.
It mentions the use of shell scripts like scripts/check-prompts.sh and scripts/run-behavioral-evals.sh for triggering evaluations.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8) because it processes untrusted user data to generate executable instructions for subagents.
Ingestion points: User-provided test prompts, conversation history, and user feedback stored in feedback.json.
Boundary markers: No specific boundary markers or delimiters are defined to separate user input from agent instructions.
Capability inventory: The skill has the ability to write files to the local filesystem (safety: writes-files), execute Python scripts, and spawn subagents to run tasks.
Sanitization: There is no evidence of input validation, escaping, or sanitization for the data processed by the skill.
[EXTERNAL_DOWNLOADS]: The skill documentation refers to the claude-mem plugin hosted on GitHub at github.com/thedotmack/claude-mem. This is identified as a dependency for the evaluation and benchmarking functionalities.

skill-creator