The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local command execution for core developer workflows.
scripts/run_eval.py uses subprocess.Popen to invoke the claude CLI to perform evaluations on skill descriptions.
eval-viewer/generate_review.py executes lsof via subprocess.run to manage local network ports for the evaluation viewer.
[EXTERNAL_DOWNLOADS]: The evaluation viewer UI (eval-viewer/viewer.html) references an external JavaScript library (SheetJS) from a public CDN (cdn.sheetjs.com). This is a well-known service used for processing spreadsheet data within the browser.
[PROMPT_INJECTION]: The benchmarking and evaluation logic in scripts/run_eval.py represents an indirect prompt injection surface. It ingests test queries from user-defined evaluation sets and passes them to an AI agent. Maliciously crafted queries in the evaluation data could potentially influence the agent's behavior during the test.
Ingestion points: Evaluation queries are loaded from JSON files (e.g., eval_set.json) in scripts/run_eval.py and scripts/run_loop.py.
Boundary markers: Absent; queries are passed as raw strings to the CLI arguments.
Capability inventory: The skill can execute local commands (claude, lsof), read and write local files, and start a local HTTP server.
Sanitization: None; queries are processed as provided by the user and interpolated into command-line arguments.

skill-creator