project-knowledge
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data, including source code, manifests, and existing documentation, to identify implementation drift and generate guides.
- Ingestion points: Accesses codebase content such as route definitions, ORM models, environment variable reads, and dependency manifests as described in references/preparation.md and references/documentation-audit.md.
- Boundary markers: Utilizes a 'file/audience plan' and 'authorized mode' (audit-only, plan-only, fix, commit) to restrict operations and scope.
- Capability inventory: Capable of writing to project files and dispatching specialized sub-tools for codebase discovery.
- Sanitization: Incorporates a specific auditing step for instruction claims using an external auditor tool and employs a reviewer for cross-document validation.
- [NO_CODE]: The skill consists entirely of Markdown-based instructions and methodology references without any executable scripts, binaries, or automated installation steps.
Audit Metadata