knowledge-promotion-review
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process 'evidence from project work, documentation, or repeated agent friction', which constitutes untrusted external data. This creates a surface for indirect prompt injection if malicious instructions are embedded in these sources to manipulate the promotion verdict or influence the agent's behavior during testing.
- Ingestion points: Project work, documentation, and agent friction reports processed in SKILL.md.
- Boundary markers: The skill requires identifying source owners, local problem context, and non-applicability boundaries, which helps delineate trusted vs. untrusted context.
- Capability inventory: The skill performs read-only reviews by default but mentions potential mutation of destinations and hand-offs to other tools like 'knowledge-handoff' and a 'promotion playbook' if explicitly authorized by the user.
- Sanitization: Instructions mandate stripping volatile local facts from reusable claims and using independent behavioral tests with fresh evaluators to verify guidance before it is promoted.
Audit Metadata