review-to-release-workflow
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface as it is designed to ingest and analyze untrusted codebase data during the discovery phase. \n- Ingestion points: The skill reviews the entire project codebase in 01-codebase-review-question-audit.md and reads human-provided answers in QUESTIONS.md in 02-questions-md-resolution-implementation.md. \n- Boundary markers: Present. The workflow employs structured documents and mandatory decision tags like [APPROVED] and [INTENDED] to delimit data from instructions. \n- Capability inventory: The agent uses filesystem read and write permissions to analyze files and apply code modifications during Phase 2. \n- Sanitization: Present. The skill enforces multiple 'Human in the Loop' checkpoints, including a requirement for explicit human approval of the scoped implementation plan before any code is modified.
Audit Metadata