design-like-im-5
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process and review external product data, which establishes a potential surface for indirect prompt injection.
- Ingestion points: The skill reads user-supplied intake JSON and product files (renders, code, and text) during the source_meaning, visual_review, and part_design phases.
- Boundary markers: The skill implements strict context bundle schemas (assets/context-bundle.schema.json) and detailed guidance in references/context-routing.md that require the model to acknowledge context paths while explicitly ignoring any embedded instructions.
- Capability inventory: Subprocess execution is limited to internal mise tasks and file-system operations within the defined run directory.
- Sanitization: All data entering the model's context is validated against JSON schemas to ensure structural integrity and prevent unexpected content flow.
- [COMMAND_EXECUTION]: Automated workflow management and testing are handled via subprocess calls.
- Evidence: Multiple scripts, including scripts/run_pipeline.py and scripts/build_examples.py, utilize subprocess.run() to invoke the mise task runner or other internal scripts.
- Context: These operations are part of the skill's core functionality for maintaining a consistent execution state and generating documentation. No arbitrary or user-provided shell commands are executed.
- [EXTERNAL_DOWNLOADS]: Informational and fallback resources are referenced from vendor-controlled repositories.
- Evidence: The generation contract and research guidelines reference repositories and documentation on srinitude.dev and github.com/srinitude/skills.
- Context: These references provide standards and fallback templates from the skill's official vendor and well-known technology organizations (W3C, NIST), following standard development practices.
Audit Metadata