The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npm install and npm run verify. These are standard operations for initializing and testing a local Node.js project. The dependencies listed in package.json are well-known, legitimate software libraries (Astro, Starlight, React, Mermaid).
[COMMAND_EXECUTION]: The skill uses basic shell commands such as cp -R for scaffolding files and find to locate the skill's own asset directory. These are necessary for the skill's stated purpose of project creation.
[DYNAMIC_EXECUTION]: The evaluation script eval/run_eval.py utilizes subprocess.run to execute a local shell script (dry-run-replay.sh) as part of a test harness. This is a standard practice for development and testing of the skill itself and is not part of the primary user-facing workflow.
[INDIRECT_PROMPT_INJECTION]: The skill includes a specific security-positive gate, check-no-artifacts.mjs, which scans content for leaked tool-call artifacts (e.g., <parameter>, </invoke>). This helps prevent accidental command leakage or injection markers from being committed to the documentation.
[SAFE]: The site includes a client-side script in astro.config.mjs for sidebar navigation enhancement. This script is limited to benign DOM manipulations within the user's browser for UI/UX purposes.

starlight-handbook-kit