The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess to interact with system tools and the platform's CLI.
scripts/run_eval.py: Uses subprocess.Popen to execute claude -p for running evaluation queries.
eval-viewer/generate_review.py: Uses subprocess.run to call lsof for clearing local ports during server initialization.
[EXTERNAL_DOWNLOADS]: Fetches client-side libraries from well-known sources to support the evaluation viewer UI.
eval-viewer/viewer.html: Downloads the SheetJS library from cdn.sheetjs.com to render spreadsheet outputs.
[DATA_EXFILTRATION]: Operates a local server to facilitate user review of test results.
eval-viewer/generate_review.py: Starts an HTTP server on 127.0.0.1 to serve the review interface. User feedback is saved to a local feedback.json file.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from external skills during its evaluation workflow.
Ingestion points: Reads transcripts and output files from the workspace directory in eval-viewer/generate_review.py and agents/grader.md.
Boundary markers: Uses role instructions for grader and analyzer subagents, but lacks explicit delimiters for processed transcript content.
Capability inventory: Includes file system access and the ability to execute platform commands via the claude CLI.
Sanitization: The viewer UI uses standard HTML escaping for displaying text content.

skill-creator