jules-implement
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by consuming output from the 'stitch-design' process and external web/paper research to drive its code implementation tasks.
- Ingestion points: The agent reads state from
.jules/ui-search-state.jsonand processes results from external web and paper searches. - Boundary markers: There are no explicit delimiters or instructions to treat external research data as untrusted content, which could allow malicious content in design hashes or search results to influence the agent.
- Capability inventory: The skill can modify the local filesystem, implement production-ready code, and create Pull Requests via the 'atomic-commit' workflow.
- Sanitization: No validation or sanitization is mentioned for the ingested design hash or search results before they are used to generate code or unit tests.
Audit Metadata