The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of building briefs and existing scene graphs, which could contain embedded instructions. The agent has significant capabilities including network access to hosted APIs and local CLI execution.
Ingestion points: Data entering the context via create_house_from_brief parameters and scene data returned by load_scene (references/tool-workflows.md).
Boundary markers: The skill encourages using semantic tools and validation (validate_scene) as defined in SKILL.md, which serves as a schema-based boundary.
Capability inventory: Network access to https://editor.pascal.app/api/mcp and local command execution via the @pascal-app/cli tool (references/setup.md).
Sanitization: The instructions mandate the use of validate_scene and verify_scene after edits, ensuring the scene graph remains within expected schemas (SKILL.md).
[COMMAND_EXECUTION]: The skill instructs the agent to install and run vendor-specific command-line tools locally.
Evidence: npm install --global @pascal-app/cli@beta and pascal editor --no-open commands in references/setup.md.
[EXTERNAL_DOWNLOADS]: The skill downloads the Pascal CLI from the NPM registry and communicates with the vendor's hosted API endpoint.
Evidence: Installation of @pascal-app/cli@beta and network connection to https://editor.pascal.app/api/mcp (references/setup.md).

pascal-3d