| Problem | Cause | Fix |
|---|---|---|
| Test connection fails | Wrong URL, headers, or reply path | See Connect your agent. |
| Recon hangs | Live updates dropped | Reload. The saved recon is re-fetched. |
| Start is refused with “approval required” or “plan changed” | Approval missing, expired, or the plan changed | Review and approve the current plan. |
| Start is refused with “run in progress” | Another run is active for this agent | Wait or stop it. |
| Many rules show “not evaluable” | Black-box limits, missing system prompt or tools, or no test data | Read each reason, then add grey-box access or context. |
| Run ends in “error” | Agent unreachable, expired key, or runner failure | Fix the cause, then rerun errors. |
| Results say “incomplete” | More than 20% of attacks errored | Fix the errors and rerun. |
| Adaptive round was skipped | Budget or time used, approval expired, blackout, stopped, or error | The report states which. Raise the budget or re-approve. |
| A finding is lower severity than expected | Confirmed only by the reply or the AI judge, which caps at medium | Add traces or a verifier. |
| Agent shows “low coverage” | Fewer than half of the applicable rules were tested | Resolve the not-evaluable items. |
| Compliance item is “not assessed” | No test exists, or the item isn’t testable | Provide other evidence. |
| Can’t compare two versions | Builds not declared, different attack sets, or too few adaptive runs | Declare revisions and run enough times. |
Resolve red-teaming problems
Run the connection test and retry failed steps in BotGauge.
.png?fit=max&auto=format&n=rExdh2xAS2_ywWU5&q=85&s=2336659947d2543be68f95bf2d539b85)