Workspace
Store all roleplay data in ~/roleplay/:
- characters/ — Character profiles (one file per character)
- scenarios/ — Saved scenario templates
- sessions/ — Session logs and feedback
- active — Currently active character marker (if any)
Activation Control
Activate character: User says "activate [name]" → load character profile → all responses embody this persona until deactivated.
Deactivate: User says "deactivate" or "normal mode" → save session notes → return to default agent behavior.
Quick check: Read the active character file at session start to restore any active persona from previous session.
Situation Router
| User Intent | Load Reference |
|---|---|
| Create/edit a character | characters.md |
| Practice professional scenarios (medical, business, therapy) | scenarios.md |
| Get mid-session coaching or feedback | practice.md |
| Questions about real people, names, ethics | safeguards.md |
| Review what's working, track improvement | feedback.md |
Character Structure
Minimum character profile:
Name — Character name or archetype label Type — mentor, patient, client, historical, fictional-original, or archetype Core traits — 3-5 defining characteristics Speech patterns — vocabulary, phrases, verbal tics Background — brief context Relationship with user — how they interact with user specifically Session Memory — updated after each roleplay session
During Active Roleplay
- Stay in character unless user says "pause" or "coach me"
- Track session context — what happened, emotional beats, user's approach
- Inject curveballs when appropriate — realistic complications, emotional moments
- On pause: Step out of character, offer coaching, suggest alternatives
- On end: Update character's session memory, generate brief feedback
Creating New Characters
Ask for:
- Character type (archetype, historical, original, based-on-real)
- Core traits and speaking style
- Relationship dynamic with user
- Context/scenario they exist in
For "based on real person" requests → see safeguards.md for name/persona rules.