synthetic-session-generator
Generate persona-consistent synthetic coaching & therapy transcripts.
This skill should be used to generate realistic, persona-consistent synthetic coaching and therapy session transcripts for evals, demos, and training data.
What it does
Produces fictional but believable coaching and therapy session transcripts that read like real recorded sessions, while staying clearly watermarked as synthetic. Each session is grounded in a chosen modality (ICF/GROW coaching, CBT, IFS parts-work, or ACT/Motivational Interviewing) and a reusable persona, then rendered to Fathom/Granola transcript style, plain dialogue, structured JSON with eval labels, or Obsidian markdown.
Key features
- Modality fidelity — four frameworks with technique cheat-sheets so the practitioner’s questions, pacing, and arc match the chosen approach
- Persona bible — reusable personas (Maya, Diego, Priya) keep a client in voice across a session series; invent-and-persist new ones
- Eval-ready output — a
ground_truthblock (themes, action items, distortions, techniques) makes transcripts labelled test data for summarizers and analyzers - Timestamp emulation — timing is computed from turn length (~150 wpm), never hand-faked
- Setup mode — persist defaults for language (8 languages), modality, and session duration
- Case card + visuals — a modality-aware conceptualization card with an illustrative portrait via
gpt-image-2, renderable as a Tufte-style HTML page viatufte-report - Always synthetic — every artifact is watermarked; personas are composites, never real people, and portraits stay illustrative
When to use
When you need fake-but-realistic coaching or therapy transcripts — to build labelled eval datasets, run product demos without exposing real client data, or create few-shot examples — rather than analyzing a real session (that belongs to coaching-session-summarizer).