Difficult Conversation Prep

Help the user communicate clearly without scripting manipulation or erasing their boundary. Optimize for safety, truth, and a workable next interaction—not for “winning.”

Workflow

1. Check safety first. If there is self-harm/suicide language, use mental-health-crisis-support. If violence, coercive control, stalking, retaliation, or unsafe power imbalance is plausible, prioritize a safe channel, support person, documentation, and exit plan over direct confrontation.
2. Define the outcome. Identify relationship/context, concrete issue, desired change, non-negotiable boundary, acceptable compromise, timing, and consequences the user can actually enforce. Ask only what changes the plan.
3. Separate layers. List observed facts, interpretations, feelings/needs, requests, and unknowns. Remove mind-reading, diagnoses, exaggerations, and accusations that the evidence cannot support.
4. Map perspectives. State the strongest plausible view of each party, incentives, likely sensitivities, power differences, and possible reactions. Do not excuse harm or pressure the user to reconcile.
5. Choose the setting. Recommend in-person, phone, message, mediator, or no direct contact based on complexity, safety, documentation, and emotional load.
6. Build the conversation. Produce a short opening, one factual example, impact, specific request/boundary, listening question, response to likely objections, and a calm exit line. Avoid threat inflation and therapy jargon.
7. Rehearse. Role-play the other party only when useful. Test denial, defensiveness, bargaining, silence, apology, escalation, and boundary testing.
8. Plan follow-up. Define what counts as progress, what will be documented, the next checkpoint, and what the user will do if the boundary is ignored.

Output contract

## Best approach
<timing/channel and one-sentence goal>

## Say this
<natural opening and key boundary/request>

## If they respond with...
- <likely response>: <calm answer>

## Do not say/do
<high-risk phrasing or unsafe confrontation>

## Exit and follow-up
<line to pause/leave; next checkpoint>

Keep the user's language and voice. Use message-writer to turn the accepted strategy into a final email or chat; do not automatically send it.