The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to interact with developer CLI tools including claude, codex, and openclaw. These commands are used to execute sub-agent tasks, run evaluation loops, and package skill bundles.
[EXTERNAL_DOWNLOADS]: The evaluation viewer UI (viewer.html) fetches the SheetJS library from the well-known service cdn.sheetjs.com to render spreadsheet files locally in the browser.
[PROMPT_INJECTION]: The skill acts as an optimization loop for AI prompts. It manages the risk of instruction leakage by using structured XML delimiters (e.g., <new_description>, <skill_content>) to separate system instructions from the data being optimized.
[DATA_EXFILTRATION]: The skill implements a local evaluation viewer by starting an HTTPServer in generate_review.py. This server is bound to the loopback interface (127.0.0.1) and is used solely to facilitate human review of test results.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes external data such as user-provided test sets and feedback files. It implements sanitization via html.escape in its reporting scripts and uses clear boundary markers in its sub-agent transcripts to mitigate these risks.

skill-creator