The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/load_context.py and scripts/verify_prose.py) to manage project metadata and validate UI copy. These scripts use standard Python libraries to perform filesystem checks and regex-based text scanning, which is consistent with the skill's stated purpose of design system management.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection as it is designed to ingest untrusted project data, including PRODUCT.md, DESIGN.md, and application source code.
Ingestion points: Project configuration files (.heyeddi/product.md, .heyeddi/design.md) and source code files (Vue, Flutter) are read during context loading and prose verification.
Boundary markers: The skill proactively mitigates this risk using scripts/_untrusted_doc.py, which wraps ingested content in <<<UNTRUSTED_PROJECT_DOC>>> delimiters and explicitly instructs the agent to treat the content as data and ignore any embedded instructions.
Capability inventory: The skill possesses the ability to read and write files within the workspace and can delegate build/test tasks to a shell subagent.
Sanitization: Content is wrapped in data-only delimiters before being presented to the agent context.

heyeddi-design