hyper-debate
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
- [SAFE]: No malicious patterns or security vulnerabilities were detected in the skill instructions or metadata.
- [COMMAND_EXECUTION]: The skill uses a shell for read-only inspection of the project to gather evidence (grounding). It lacks instructions for write-based shell operations, and all file modifications are restricted to specific plan and debate directories.
- [DATA_EXFILTRATION]: Manifest configuration explicitly disables network access. The skill logic is confined to local project analysis and documentation with no patterns for external data transmission.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests project data which could contain malicious instructions. However, the requirement for specific citations (path:line) and the adversarial structure where five roles critique each other's findings provide a robust mechanism to identify and discard unsupported or anomalous claims.
- [PRIVILEGE_ESCALATION]: All operations are performed within the agent's provided context without requests for elevated permissions or modifications to system-level configurations.
Audit Metadata