The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires and provides internal scripts to manage its 3D optimization and verification workflow.
Evidence: scripts/shot-server.py is a Python-based HTTP server launched by the agent to receive screenshot data from the browser and save it to the disk.
Evidence: scripts/compare.sh is a shell script that invokes ImageMagick tools to generate visual comparison montages and pixel-level diffs.
Context: These tools are intended for use in a local development environment (the 'studio' page).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project-level data that guides the agent's remodeling decisions, creating an attack surface for indirect prompt injection.
Ingestion points: The agent reads instructions from a project's 3D-STYLE.md file (Step 0 in SKILL.md) and project configuration data in package.json (Step 1 in references/setup.md).
Boundary markers: There are no explicit delimiters or specific warnings provided to the agent to treat content within these project files as non-authoritative data.
Capability inventory: The agent possesses capabilities to execute shell scripts, write files, and perform browser-based execution (javascript_tool), which could be leveraged if the 3D-STYLE.md contains malicious instructions.
Sanitization: No sanitization or validation is performed on the content of the project's style guide or configuration files before processing.

zk-model-pass