The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently uses the subprocess module to interact with the local system. scripts/run_eval.py and scripts/improve_description.py execute the claude CLI tool using arguments derived from user-provided queries and skill content. Furthermore, eval-viewer/generate_review.py executes the lsof command and calls os.kill to manage network ports, which could be misused to terminate arbitrary processes if the port inputs are not strictly controlled.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection (Category 8) when processing user-defined test cases and queries. Ingestion points: Test prompts are read from evals/evals.json and external eval sets. Boundary markers: The queries are interpolated directly into prompts for the claude CLI without protective delimiters or instructions to ignore embedded commands. Capability inventory: The skill possesses powerful capabilities including subprocess execution, filesystem access (e.g., writing to ~/.claude/commands/), and skill management via host.skills. Sanitization: The skill lacks sanitization for the user-provided queries before they are passed into the evaluation and optimization workflows.
[EXTERNAL_DOWNLOADS]: The eval-viewer/viewer.html file references a library from cdn.sheetjs.com. This is a well-known service utilized by the skill for rendering and manipulating spreadsheet data within the results viewer.

skill-creator