The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and review user-supplied artifacts such as screenshots, captures, and source code. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data being reviewed.
Ingestion points: Processes external code, screen captures, and design states (referenced in SKILL.md and review-playbook.md).
Boundary markers: The skill mandates a clear distinction between 'observations' (factual data from evidence) and 'hypotheses' (inferred intent), reducing the risk of the agent following unintended instructions in the source material.
Capability inventory: Accesses tool calls for status recording (omh runtime record) and delegates tasks to other specialized workflow lanes (e.g., frontend, accessibility-audit).
Sanitization: While no specific input sanitization is mentioned, the instructions explicitly restrict the agent to preparing briefs and findings rather than executing the analyzed code.
[EXTERNAL_DOWNLOADS]: The skill references external GitHub repositories for technical research, including GSAP and other web-based UI libraries. The documentation in web-production-libraries.md explicitly warns that the platform does not fetch or install these libraries at runtime and requires the implementation owner to conduct separate license and security reviews before any manual integration.
[SAFE]: All primary technical guidance is sourced from trusted official domains (developer.apple.com, w3.org). The skill emphasizes adherence to native platform security controls and accessibility standards, while maintaining a strict policy against representing generated or prepared content as certified or implementable without observed evidence.

omh-apple-design