The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module and os.kill to execute benchmarking scripts and manage the lifecycle of a local results viewer. These operations are targeted at managing its own server and running local evaluation tasks.
[EXTERNAL_DOWNLOADS]: Interacts with the Anthropic API via the anthropic Python SDK to facilitate AI-assisted optimization of skill descriptions. This communication is restricted to official endpoints and is a necessary feature of the skill.
[PROMPT_INJECTION]: As a tool designed to ingest user prompts and feedback to generate instructions for other AI agents, the skill naturally possesses an indirect prompt injection surface.
Ingestion points: User prompts in SKILL.md, iterative feedback stored in feedback.json, and test prompts in evals.json.
Boundary markers: The skill uses standard Markdown and YAML structure but does not include explicit delimiters or safety instructions within the data it processes to prevent adversarial overrides.
Capability inventory: The skill possesses the capability to write files, execute local commands via subprocess, and make network requests through the Anthropic API.
Sanitization: Implements structure and length validation in quick_validate.py and uses yaml.safe_load to mitigate common deserialization risks.

skill-creator