The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_eval.py uses subprocess.Popen to invoke the claude CLI. This is used to test whether the agent triggers the skill correctly for a given description, which is core to the skill's purpose.
[EXTERNAL_DOWNLOADS]: The scripts scripts/improve_description.py and scripts/run_loop.py utilize the anthropic Python library to interact with the Anthropic API. This is necessary for the automated description optimization loop powered by the LLM.
[COMMAND_EXECUTION]: In eval-viewer/generate_review.py, the skill executes lsof and kill via subprocesses to manage the local web server port. This is a standard utility function to ensure the viewer can start on the requested port.
[DATA_EXFILTRATION]: The eval-viewer tool reads files from the local evaluation workspace and embeds them into a self-contained HTML file for user review. This processing occurs locally and the resulting server binds to 127.0.0.1 by default, ensuring the data remains on the user's machine.
[PROMPT_INJECTION]: The skill utilizes specialized subagents (grader.md, comparator.md, analyzer.md) to evaluate outputs. While these subagents ingest data from skills being tested, they are provided with structured instructions and schemas to mitigate risks associated with indirect prompt injection from the tested content.

skill-creator