ulw-qa
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a structured workflow for testing and verification without utilizing any dangerous commands, unauthorized network access, or persistence mechanisms. Its logic is entirely focused on task coordination and reporting within the agent's platform.
- [INDIRECT_PROMPT_INJECTION]: The skill implements a data ingestion surface for processing results from external adversarial 'probes'. While these inputs are potentially untrusted, the skill incorporates significant architectural mitigations to prevent accidental instruction execution.
- Ingestion points: Data is ingested via the
resultfield of probe tasks, as described inreferences/board-fanin.md. - Boundary markers: The workflow utilizes separate execution lanes, mandates isolated worktrees for destructive testing, and requires explicit user 'go-ahead' signals before advancing phases.
- Capability inventory: The skill uses platform-native tools for task management (
omh_agent_board), internal routing (omh_delegate_route), and isolated environment creation (worktrees). - Sanitization: The workflow relies on manual re-verification and human review gates in the review lane before findings are treated as validated fixes.
Audit Metadata