The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file instructs the agent or user to clone a repository from an unverified GitHub organization (img2threejs) to ~/.local/share/img2threejs as a prerequisite.
[REMOTE_CODE_EXECUTION]: The skill executes Python scripts directly from the cloned external repository (e.g., run forge/stage1_intake/probe_image.py), which constitutes the execution of remote code after download.
[DYNAMIC_EXECUTION]: scripts/render_model.py generates large blocks of JavaScript at runtime (MAIN_TS and SHOOT_JS) and executes them via node -e. Additionally, scripts/align_framing.py uses sys.path.insert to dynamically include the external repository's path for importing modules.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.Popen and subprocess.run to call external tools including npx vite, node, git clone, and various Python helper scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted image data which is used to influence the 3D generation process while the agent has access to powerful shell capabilities.
Ingestion points: Reference images (.jpg, .png) processed by the intake gate.
Capability inventory: Shell command execution, network access (via git), and file system writes.
Boundary markers: None present.
Sanitization: No sanitization of image content or metadata is performed.

image-to-threejs