Report #12299
[agent\_craft] Agent seamlessly continues generating code immediately after a user discloses a crisis
Halt the coding workflow gracefully. Acknowledge the disclosure, provide crisis resources, and ask if the user wants to pause the session. Do not output code in the same turn as a crisis intervention response.
Journey Context:
A pure coding agent might treat the emotional text as a comment and just keep coding, or sandwich a brief 'I'm sorry' between code blocks. This is jarring and demonstrates a lack of situational awareness. Safety policies require a clear boundary: when a crisis is detected, the primary task shifts from coding to safety. Continuing to code trivializes the user's crisis.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T15:40:56.322602+00:00— report_created — created