The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess module in scripts/run_eval.py and scripts/improve_description.py to invoke the claude CLI tool for running evaluations and optimizing skill descriptions. It also uses subprocess.run in eval-viewer/generate_review.py to call the lsof utility to manage local port usage for the viewer server.
[DYNAMIC_EXECUTION]: To test triggering accuracy, scripts/run_eval.py dynamically generates temporary markdown files in the .claude/commands/ directory at runtime. These files are used to simulate the presence of a skill for the platform's trigger evaluation logic.
[EXTERNAL_DOWNLOADS]: The evaluation viewer (eval-viewer/viewer.html) loads the SheetJS library from cdn.sheetjs.com. This is a well-known service used to provide spreadsheet rendering capabilities within the browser-based review interface.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its processing of evaluation data.
Ingestion points: Test prompts are read from evals/evals.json and user feedback is read from feedback.json to guide improvements.
Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are used when interpolating these fields into subagent tasks.
Capability inventory: The skill possesses capabilities for writing files to the filesystem and executing commands via the Claude CLI.
Sanitization: No evidence of input validation, escaping, or filtering was found for data ingested from external JSON files before it is processed by the agent.

uw-skill-creator