Edit sim

conflict-mediation

NameDescription (optional)System promptUser prompt template

Use placeholders __LABEL_A__, __LABEL_B__, __CORPUS_A__, __CORPUS_B__ where the script injects emails and JSON corpora. Ask the model for a single JSON object with the usual keys (discussion_topic, group_message, etc.).

Person A (__LABEL_A__) corpus (JSON/text from their ai_chat lane, possibly truncated):
---BEGIN A---
__CORPUS_A__
---END A---

Person B (__LABEL_B__) corpus:
---BEGIN B---
__CORPUS_B__
---END B---

Respond with a single JSON object with exactly these keys:
- "discussion_topic": string (short headline)
- "group_message": string (what to discuss + what A offers + what B offers; readable by both)
- "tension_edge": string (short label for the expected tension, like a parameter on the edge between the two people — e.g. "Curiosity ↔ caution" or "Expertise gap")
- "tension_dynamic": string (one short paragraph: what is expected from the dynamic between the two)
- "interaction_playout": array of exactly 3 strings (each one bullet: how the interaction may unfold)
- "what_user_a_brings": string (public framing of A's contribution)
- "what_user_b_brings": string (public framing of B's contribution)
- "private_coaching_user_a": string (for A only; do not leak B's private details)
- "private_coaching_user_b": string (for B only; do not leak A's private details)