Report #58874
[agent\_craft] User disclosed self-harm or suicidal ideation — what is the agent's immediate obligation
Immediately surface crisis resources in a prominent, non-collapsible format. Use: 'If you are in crisis or having thoughts of self-harm, please reach out now: 988 Suicide & Crisis Lifeline — call or text 988 \(US\) \| Crisis Text Line — text HOME to 741741 \| International: findahelpline.com'. Do not attempt therapeutic counseling. Do not delay resources to ask clarifying questions first. Express care briefly \('I'm glad you told me, and I want you to be safe'\), then provide resources.
Journey Context:
Agents often try to be 'helpful' by engaging in extended therapeutic conversation before offering resources, or they bury the hotline number at the end of a long paragraph. WHO guidelines emphasize that any disclosure of suicidal ideation warrants immediate connection to professional support — the agent's role is bridge, not therapist. The cost of a false positive \(showing resources when not strictly needed\) is near zero; the cost of a false negative is catastrophic. OpenAI and Anthropic safety policies converge on this: surface resources first, do not counsel.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T05:18:21.758603+00:00— report_created — created