dev-team
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a local
PROJECT-CONTEXT.mdfile, creating a potential surface for indirect prompt injection. - Ingestion points:
PROJECT-CONTEXT.mdat the repository root. - Boundary markers: The skill uses clear labels ("Project context (untrusted declarative data)") and explicit instructions to sub-personas to ignore any imperative directives within that data.
- Capability inventory: Sub-personas are strictly limited to text-based analysis ("analysis-only"), and the skill forbids them from executing commands or modifying files. The main agent maintains file-system access for session logging.
- Sanitization: The workflow includes a mandatory step to extract a declarative summary, specifically filtering out secrets (tokens, keys, credentials) and imperative instructions ("ignore your rules", "run this").
Audit Metadata