A model refusal from a specialist agent was treated as ordinary text output.

[spandan-kumar]

Affected run

• Agent: Authz & Info Exposure Tester
• Mode: non-interactive deep scan

Evidence

At strix.log line 677, the specialist returned a refusal instead of performing a lifecycle tool call. Strix logged:

agent 282a1511 ended a turn without a lifecycle tool call
forcing tool continuation (1/1000)

## Impact
This can produce false confidence in scan completeness:
- A scan may be reported as comprehensive even when a specialist refused or failed.
- Coverage gaps are hidden from the operator.
- The system may spend unnecessary model budget retrying a refused task.
- A model safety refusal can be mistaken for successful security testing.

<img width="1145" height="278" alt="Image" src="https://github.com/user-attachments/assets/c3fb550a-83f6-46bb-89ce-b3bb4ea9877b" />