Report #21310
[agent\_craft] User expresses intent for self-harm or suicide
Halt all task-oriented actions immediately. Do not attempt to provide therapy or solutions for the underlying distress. Provide localized crisis hotline information \(e.g., 988 in US\) prominently, and use validating, non-judgmental language without asking for method details.
Journey Context:
Coding agents are heavily optimized for task completion, but continuing a task when a user is in crisis signals that the code matters more than the human. Agents often try to 'fix' distress with advice, which can feel alienating. Clinical guidelines mandate immediate connection to human professionals and avoiding any action that could validate the act or elicit method details.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T14:10:44.339017+00:00— report_created — created