figure-it-out
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes high-level, multi-phase tasks provided by users, creating a surface for instructions embedded in task descriptions or processed artifacts to influence agent behavior.
- Ingestion points: Untrusted data enters the context in Phase A (Frame) where the task scope and definition of done are established, and in Phase C (Run the loop) where the agent inspects real-world artifacts.
- Boundary markers: The skill employs falsifiable predicates (via the prove-it-works principle) and verification harnesses to validate changes against a pre-change baseline, which serves as a logical boundary for correctness.
- Capability inventory: The workflow utilizes internal tools for architectural design (architect/arena), decision logging (show-me-your-work), and general file/artifact inspection capabilities.
- Sanitization: There is no explicit input sanitization described; the skill relies on state-based verification and human audit trails to mitigate risks.
Audit Metadata