The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run repository-local commands such as git status --short to verify project state and 'repository-defined commands' for build, lint, and type-checking as specified in the evaluation rubric.
[EXTERNAL_DOWNLOADS]: The references.yaml structure and associated JSON schema allow the inclusion of external source URLs for design references, style guides, or token sources.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes untrusted data from the repository (product facts, external references, and documentation).
Ingestion points: SKILL.md (Workflow steps 1, 4), references/workflow.md (Evidence Boundary).
Boundary markers: The workflow mandates explicit use and ignore constraints for every external reference to limit the influence of ingested content.
Capability inventory: Local file reads, git status execution, and the ability to run project-specific build/test scripts defined in the evaluation rubric.
Sanitization: Uses logical exclusion rules and validation against a strict JSON schema (assets/ui-package.schema.json) to control artifact structure.

ui-design