The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and subprocess.Popen in scripts/run_eval.py and scripts/improve_description.py to execute the claude CLI. These operations are used to test skill triggering and generate description improvements. The commands are constructed using list arguments, which prevents shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes and analyzes the outputs of automated test runs, representing a potential indirect prompt injection surface.
Ingestion points: Untrusted data from execution transcripts and output files are read by the grader.md agent and the generate_review.py viewer.
Boundary markers: The skill relies on the model's instructions to distinguish task data from instructions, with few explicit delimiters used during data interpolation.
Capability inventory: The skill can execute the claude CLI, manage file systems (zipping), and host a local HTTP server.
Sanitization: HTML reports use html.escape to sanitize content before display.
[EXTERNAL_DOWNLOADS]: The evaluation viewer (viewer.html) references the SheetJS library from cdn.sheetjs.com, a well-known service, to enable the local rendering of Excel files within the review interface.

skill-creator