The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/optimize.mjs script uses a dynamic import() call with a computed URL to load a sibling inspection module from ../../threejs-model-inspect/scripts/inspect.mjs. Loading code from paths constructed at runtime is a risk factor as it depends on the integrity of the relative file system structure.
[INDIRECT_PROMPT_INJECTION]: The skill processes glTF/GLB files and extracts metadata, including node and material names, to generate a before/after report. This data is displayed to the agent, creating a surface where a maliciously crafted 3D model could influence the agent's behavior via instructions embedded in its metadata.
Ingestion points: The addressableNames function in scripts/optimize.mjs ingests node and material names from the model via the inspect helper.
Boundary markers: None. The extracted names are printed directly to the agent's console output in a comma-separated list without delimiters or warnings.
Capability inventory: The skill has the capability to execute shell commands (spawnSync) and perform file system operations.
Sanitization: None. While the script filters out default names (e.g., material_1), it does not sanitize names for natural language instructions or prompt injection patterns.
[COMMAND_EXECUTION]: The script executes external commands via spawnSync to run the gltf-transform utility. On Windows platforms, the script explicitly enables shell: true to invoke npx. If an attacker manages to place a file with a malicious name (e.g., containing shell metacharacters) and bypasses the stat() check, it could lead to command injection.
[EXTERNAL_DOWNLOADS]: The script uses npx to fetch and execute the @gltf-transform/cli@4 package from the NPM registry. This is a standard operation for a well-known service used to provide the skill's primary functionality.

threejs-model-optimize