The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design inputs such as screenshots and wireframe markdown files that could contain malicious instructions designed to influence code generation or agent behavior.\n
Ingestion points: scripts/load_handoff.py and scripts/describe_handoff.py identify desktop.png, mobile.png, wireframe.md, and mockup-brief.md as primary inputs from the workspace.\n
Boundary markers: The skill implements scripts/_untrusted_doc.py to wrap ingested content in <<<UNTRUSTED_PROJECT_DOC>>> delimiters, accompanied by explicit warnings to the agent to treat the content as data only and ignore embedded instructions.\n
Capability inventory: The skill is capable of writing Vue components (.vue), CSS style sheets (tokens.css), and updating project documentation (design.md) and logs.\n
Sanitization: The skill relies on the aforementioned structural delimiters and clear agent-level instructions in SKILL.md and context/ANTI_PATTERNS.md to enforce data/instruction separation.\n- [COMMAND_EXECUTION]: The skill invokes several local Python scripts to manage the design-to-code workflow and verify implementation accuracy.\n
Evidence: manifest.json defines tools that execute scripts/load_handoff.py, scripts/describe_handoff.py, scripts/verify_handoff.py, scripts/verify_tokens.py, and scripts/verify_theme.py.\n
Context: The workflow in SKILL.md also orchestrates calls to scripts in sibling vendor skills (e.g., engineering-excellence, heyeddi-design) for engineering audits and prose validation.

heyeddi-handoff