🎬 STUDIO
AI that plans before it builds, learns from mistakes, and never forgets.
Quick Start •
How It Works •
Commands •
Scripts •
Full Guide
The Problem
Most AI coding assistants:
- ❌ Assume requirements instead of asking
- ❌ Declare success without verification
- ❌ Forget your preferences between sessions
- ❌ Repeat the same mistakes
STUDIO fixes this.
🚀 Quick Start
# Install
/plugin marketplace add https://github.com/twofoldtech-dakota/studio.git
/plugin install studio@twofoldtech-dakota
# Plan something
/studio "Add user authentication with email verification"
# Build the plan
/build task_xxx
That's it. STUDIO asks questions, creates a plan, and executes with validation.
🔄 How It Works
YOU STUDIO OUTPUT
│ │ │
│ "Add auth" │ │
├───────────────────────►│ │
│ │ │
│ ◄── Questions ───────┤ (3 rounds) │
│ ─── Answers ────────►│ │
│ │ │
│ ◄── Plan ────────────┤ (review & approve) │
│ ─── "looks good" ───►│ │
│ │ │
│ ├───────────────────────►│ Code
│ │ Build + Validate │ Tests
│ │ Quality Gates │ Docs
│ │ Learn & Remember │
│ │ │
The Pipeline
| Phase | What Happens | Script |
|---|
| Plan | Questions → Requirements → Steps | confidence-score.sh |
| Validate | Structure check, confidence ≥70 | validate-plan.sh |
| Pre-check | Lint, types, existing issues | quality-precheck.sh |
| Build | Execute steps, track progress | step-progress.sh |
| Verify | Acceptance criteria, DoD | verify-ac.sh, dod-check.sh |
| Learn | Capture patterns, evolve knowledge | sprint-evolution.sh |
📋 Commands
| Command | Alias | What it does |
|---|
/studio "goal" | /s | Start planning with questions |
/build task_xxx | /b | Execute an approved plan |
/build --resume | | Continue from last step |
/status | | Check current task state |
See docs/QUICK-REFERENCE.md for all commands.
🔧 Scripts
All scripts output JSON, have --help, and live in scripts/.
Quality Pipeline
# Before build
./scripts/validate-plan.sh --task-id task_xxx # Structure check
./scripts/confidence-score.sh --task-id task_xxx # Quality score (0-100)
./scripts/quality-precheck.sh # Lint + typecheck
# During build
./scripts/step-progress.sh status task_xxx # Track progress
./scripts/error-matcher.sh --input "error text" # Fix suggestions
# After build
./scripts/verify-ac.sh --task-id task_xxx # Acceptance criteria
./scripts/dod-check.sh --auto-detect # Definition of Done
Confidence Scoring
Plans are scored 0-100:
| Category | Points | Checks |
|---|
| Requirements | 25 | User confirmations, edge cases, scope |
| Step Quality | 25 | Success criteria, atomic actions |
| Context | 25 | Constraints, quality requirements |
| Risk | 25 | Retry behavior, failure handling |
≥85 = PROCEED • 70-84 = CAUTION • <70 = BLOCKED
🧠 Knowledge System
STUDIO learns from every build:
| Section | Purpose |
|---|
| Strict Constraints | Rules that must never be violated |
| Slop Ledger | Naming/structural mistakes to avoid |
| Performance Delta | Measured improvements with metrics |
| Pending Queue | Signals awaiting promotion |
Every 5 tasks → automatic evolution proposals.
📁 Project Structure