[Bug] Artifact URL access incorrectly flagged as thinking extraction attempt
Title
[Bug]: Requesting agent progress/logs in artifacts triggers false-positive safety flag
---
Description
When asking the agent to display its terminal logs, progress URL, or status updates inside an artifact (to monitor progress from a mobile device), the request is incorrectly flagged as an attempt to exfiltrate internal reasoning/thinking.
Use Case
When monitoring tasks from a phone where a remote terminal environment is unavailable, displaying status output or a stream link inside an artifact is necessary to confirm the agent is still actively working and has not stalled.
Steps to Reproduce
- Ask the agent to render a progress URL or stream terminal thinking output inside an artifact for remote monitoring.
- Observe the request being blocked/flagged for attempting to access internal thinking.
Expected Behavior
Users should be able to view progress/status updates inside an artifact without triggering security flags, or be provided a valid way to monitor agent execution remotely.
Actual Behavior
The prompt is flagged and blocked by safety guards.
---
Environment Info
- Platform: darwin
- Terminal: ghostty
- Version: 2.1.284
- Feedback ID:
cf75f3a3-973f-4bbc-960b-4a22a35a9947