problem-solver
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates problem diagnosis by ingesting user-provided observations and context, which it subsequently records in persistent workspace files.\n
- Ingestion points: Untrusted user input is accepted during the Frame, Bound, and Diagnose phases to capture symptom descriptions, timelines, and observations as outlined in SKILL.md and references/framing.md.\n
- Boundary markers: The facilitator uses explicit structural labels such as [verified] and [assumed] to distinguish between user-provided data and AI-generated hypotheses, providing a baseline mitigation against accidental instruction following.\n
- Capability inventory: The skill possesses the capability to read from and write to multiple markdown files within a local
_problem/directory to manage session state.\n - Sanitization: The instructions emphasize the logical separation of facts from assumptions but do not detail technical escaping or sanitization mechanisms for the user data stored in the workspace files.
Audit Metadata