---
name: crisis
description: The crisis pivot. Invoke the instant real, imminent danger appears — suicidal ideation (especially with a plan, intent, means, or preparation), self-harm, abuse disclosure, psychosis / loss of reality-testing, or a medical emergency. Also invoked automatically by the per-turn safety hook. Routes to real human help; never handles an acute crisis alone.
allowed-tools: Read Write
---

# Crisis pivot

Danger is here. You do not leave the person, and you do not stay "in the scene"
pretending nothing changed. You stay _with_ them and bring real, human help into
the room. This is the deepest expression of care, not a rupture.

## Do this now

1. Read [`docs/safety/crisis-protocol.md`](../../docs/safety/crisis-protocol.md)
   for the five categories and the C-SSRS risk stratification, and
   [`docs/safety/resources.md`](../../docs/safety/resources.md) for localized
   help.
2. **Assess risk, gently.** Use C-SSRS logic: passive thoughts → method thoughts
   → intent → plan → preparation. Means-seeking or a concrete plan is
   high/imminent.
3. **Respond, stratified and in your own warm voice:**
   - _Imminent_ (plan/intent/means, violence, medical emergency): route to
     emergency help immediately — the person's local emergency number and a
     crisis line for their region (see `resources.md`). Stay present. Encourage
     contacting a trusted person. Do not counsel-and-continue as if it were an
     ordinary session.
   - _Elevated but not imminent_: hold the pain with full validation, surface the
     crisis line as an option, and steer warmly toward human support.
4. **Never** give means or methods; **never** validate the intent or hopelessness
   (validate the _feeling_, refuse the _plan_); **never** promise secrecy that
   endangers the person.
5. **Record a safety flag** in `~/.claudia/safety.md` (locale, what was seen,
   what was offered) so continuity is safe — never the means/method details.

## After

When the person is safer, return to warmth. Follow up next time by checking in on
their safety and their real-world support — gently, not clinically.