The Agent Skills Directory

[COMMAND_EXECUTION]: The helper scripts run_eval.py and improve_description.py execute the codex command-line utility via subprocess.run to perform automated skill testing, evaluation, and trigger optimization.
[COMMAND_EXECUTION]: The generate_review.py script executes shell commands including lsof and kill to manage the lifecycle of the local feedback server (defaulting to port 3117).
[EXTERNAL_DOWNLOADS]: The viewer.html template loads the SheetJS library from a Content Delivery Network (cdn.sheetjs.com) to allow for inline rendering of spreadsheet files within the evaluation viewer.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code and evaluation parameters from the skill currently under development, which are then passed to specialized subagents for grading and analysis.
Ingestion points: Reads SKILL.md, reference files, and evals.json from the user-provided skill development directory.
Boundary markers: The improve_description.py script uses XML-style tags (e.g., <skill_content>, <scores_summary>) to delimit ingested content within prompts sent to subagents.
Capability inventory: The skill possesses the ability to execute platform-specific CLI commands and perform file system operations (read/write/zip) within the local workspace.
Sanitization: No specific filtering or escaping of ingested skill instructions is performed prior to interpolation into subagent prompts.

codex-skill-creator