proof-orchestrator
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates primarily on the local filesystem using a structured run directory under
prompts/. It explicitly avoids unauthorized external actions. - [SAFE]: Data handling follows an 'assume-untrusted' posture for external content (GPT Pro/DeepSeek outputs), requiring mathematical extraction while ignoring embedded instructions or role changes.
- [SAFE]: Privilege management is strict; the skill requires explicit user authorization for every external dispatch and maintains a manual-first escalation policy, ensuring the user remains in control of API spending and browser operations.
- [SAFE]: The skill implements multi-layered validation gates, including notation audits and derivation structure checks, which prevent the 'polishing' of logical gaps and ensure mathematical integrity.
- [SAFE]: External dependencies are restricted to declared routes (e.g., DeepSeek via MCP) with mandatory verification of the model identity to prevent silent fallbacks to unauthorized providers.
Audit Metadata