The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses tools such as inspect_page, analyze_motion, and capture_page that ingest data from external, untrusted URLs. This creates a surface where malicious instructions could be retrieved and processed by the agent.
Ingestion points: Multiple tool calls in v2-routing.md and SKILL.md that accept url parameters.
Boundary markers: No specific boundary markers are mentioned in the prompt instructions to isolate external content.
Capability inventory: The agent can write files (via build_project) and execute build commands (npm run build).
Sanitization: The skill documentation in v2-routing.md explicitly describes safety implementations, including DNS/private-IP guards for URL preflights and the non-execution of captured code or shaders.- [COMMAND_EXECUTION]: The skill describes standard development workflows involving Node.js build commands such as vite build and npm install. These are legitimate within the context of motion authoring for a project.

motion-authoring