A structured Cognitive Behavioral Therapy toolkit that turns your AI assistant into a guided thought record companion. 12-step records, daily check-ins, and pattern analysis — all stored locally.
cbt-llm-kit provides structured CBT (Cognitive Behavioral Therapy) thought recording through your AI assistant. Based on the Beck/Burns cognitive model, it guides you through identifying automatic thoughts, examining evidence, recognizing cognitive distortions, and building alternative perspectives.
Your AI reads the schema, questions, and cheat sheet — then walks you through each step, one question at a time. Records are saved as local JSON files for privacy.
Run the installer from any project directory where you want the commands available:
curl -sL https://raw.githubusercontent.com/arktnld/cbt-llm-kit/main/install.sh | bashThe installer will ask which AI you use, then set up data/, records/, and the slash commands in your current directory.
Open your AI assistant in the project and run:
/cbt:record
The AI will guide you through all 12 steps:
AI: What happened? Where, when, and with whom?
You: I was at the grocery store. The security guard kept watching me
and the cashier acted strange.
AI: What went through your mind at that moment?
You: It's my fault — as if I had done something wrong.
AI: What did you feel? Which emotion and what intensity from 0 to 100%?
You: Discomfort at 50, rejection at 70.
...continues through all 12 steps...
After recording one or more situations, reflect on the day:
/cbt:checkin
The AI reads your records, summarizes them, and walks you through identifying patterns — common distortions, recurring emotions, and safety behaviors.
After several days of recording:
/cbt:analyze
The AI reads all your records and identifies trends: most frequent distortions, how emotion intensity changes over time, recurring safety behaviors, and emerging core beliefs.
| Assistant | Support | Command Format |
|---|---|---|
| Claude Code | ✅ | /cbt:record |
| Gemini CLI | ✅ | /cbt_record |
| Cursor | ✅ | /cbt:record |
More AI assistants coming soon.
Each thought record follows this structured sequence:
| Step | Field | Question |
|---|---|---|
| 1 | Situation | What happened? Where, when, and with whom? |
| 2 | Automatic thought | What went through your mind? |
| 3 | Emotions (before) | What did you feel? Intensity 0-100% |
| 4 | Physical sensations | What did your body feel? |
| 5 | Behavior | What did you do at the time? |
| 6 | Evidence for | What supports this thought? |
| 7 | Evidence against | What contradicts this thought? |
| 8 | Alternative thought | A more balanced way to see it? |
| 9 | Safety behavior | Anything you do to avoid anxiety? |
| 10 | Cognitive distortions | Which distortion is present? |
| 11 | Core beliefs | Deeper belief about yourself? |
| 12 | Emotions (after) | How do the emotions feel now? |
The schema includes all 13 Beck/Burns cognitive distortions:
| # | Distortion | Example |
|---|---|---|
| 1 | Mind reading | "He looked at me because he suspects me" |
| 2 | Personalization | "It's my fault" |
| 3 | Catastrophizing | "Everything will go wrong" |
| 4 | Selective abstraction | "One person criticized, so it was bad" |
| 5 | Arbitrary inference | "Nobody called, they don't like me" |
| 6 | Overgeneralization | "I failed once, I always fail" |
| 7 | Magnification/minimization | "My mistake was huge, my success was luck" |
| 8 | All-or-nothing thinking | "If it's not perfect, it's a failure" |
| 9 | Fortune telling | "The interview will be a disaster" |
| 10 | Emotional reasoning | "I feel like a failure, so I am one" |
| 11 | Labeling | "I'm useless" |
| 12 | Should statements | "I should be stronger" |
| 13 | Disqualifying the positive | "I passed, but it was luck" |
| File | Description |
|---|---|
data/schema.json |
Record structure with all valid field values — 13 distortions, 19 emotions, 14 physical sensations, 3 core belief categories |
data/questions.json |
12-step record flow + 3-step daily check-in |
data/cheat-sheet.md |
Complete CBT reference — cognitive model, distortions table, questioning techniques, exposure hierarchy |
All records are stored locally in records/. No data is sent to external services.
Inspired by spec-kit and its approach to AI-agent-agnostic tooling.
This project is licensed under the terms of the MIT open source license. See LICENSE for details.