plan-roadmap
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes multiple external and potentially untrusted project files to generate roadmaps, creating an attack surface where instructions embedded in project specifications could influence agent behavior.
- Ingestion points: The skill reads
docs/specifications/product.md,AGENTS.md,CONTEXT.md, architecture docs, ADRs, and the existing codebase. - Boundary markers: Absent. There are no instructions to use delimiters or to treat content from these files as untrusted data.
- Capability inventory: The skill has the ability to write to
docs/planning/roadmap.mdanddocs/planning/project-status.mdbased on the ingested content. - Sanitization: Absent. The process extracts information directly from input documents and incorporates it into planning files without explicit filtering or validation of embedded commands.
Audit Metadata