The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The generated Three.js viewer file (mannequin.html) uses an ESM import map to fetch the Three.js library and its addons from the well-known unpkg.com CDN. This is a standard practice for web-based visualizations and the source is a recognized service.
[COMMAND_EXECUTION]: The skill includes a standalone script, scripts/skeleton.mjs, which is used to process joint math and generate the viewer. It uses the node:fs module to read pose data and write the resulting HTML file. These operations are restricted to the local file system and are consistent with the skill's primary purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON data for poses and animations. It includes a robust validation layer in scripts/skeleton.mjs that checks all joint and movement names against hardcoded allowlists (JOINTS and MOVEMENTS) before processing numeric values. This effectively prevents malicious instructions from being executed or influencing the agent's logic through data poisoning.
Ingestion points: Reads .json files via the --file flag or raw strings via the --cmd flag in scripts/skeleton.mjs.
Boundary markers: The script uses JSON.parse and strictly validates the resulting object structure.
Capability inventory: File read/write capabilities are limited to pose files and the HTML viewer template.
Sanitization: All movement names are validated against an internal allowlist; unknown or invalid names cause the script to exit with an error.
[REMOTE_CODE_EXECUTION]: While the skill loads remote JavaScript in the browser environment via the viewer, this is limited to the Three.js library. The skill itself is zero-dependency and does not perform any remote code execution or package installation on the host system.

octocode-mannequin