The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess module to interact with the system environment and the platform's command-line interface.
scripts/improve_description.py and scripts/run_eval.py execute the claude CLI to perform sub-agent evaluations and test skill triggering. This is a documented interaction pattern for this platform.
eval-viewer/generate_review.py uses the lsof utility to detect and manage conflicts on local network ports.
[EXTERNAL_DOWNLOADS]: The evaluation viewer UI references an external JavaScript library for data processing.
eval-viewer/viewer.html loads the xlsx library from cdn.sheetjs.com. This is a well-known service providing common utility libraries.
[DATA_EXFILTRATION]: The skill starts a local HTTP server to provide a user interface for reviewing test results.
eval-viewer/generate_review.py runs an HTTP server on 127.0.0.1 (default port 3117). This activity is confined to the local loopback interface and is intended for user interaction with the benchmark data.
[PROMPT_INJECTION]: The skill ingests user-provided prompts and feedback to refine agent instructions.
While files like evals.json and feedback.json create a surface for indirect prompt injection in the evaluation loop, this is a necessary part of the skill's primary function as a developer tool. The risk is mitigated by the iterative, human-in-the-loop design of the workflow.

skill-creator