PromptTrace

Local-first prompt optimization and reuse engine for AI-assisted development.

Scans your AI coding tool histories, classifies every prompt, scores quality, extracts reusable templates, detects workflow patterns, and gives you a dashboard to improve your prompting. All data stays on your machine.

Quick Start

npx prompttrace

That's it. Opens your browser with the dashboard. First run takes ~60s to install dependencies and build. After that, starts in ~3s.

Also works with:

# yarn
yarn dlx prompttrace

# pnpm
pnpx prompttrace

# bunx
bunx prompttrace

# with custom port
npx prompttrace --port 4000

# scan only (no dashboard)
npx prompttrace scan

# skip auto-scan
npx prompttrace --no-scan

What It Does

PromptTrace reads local history files from your AI coding tools, then:

1. Classifies every prompt into 14 categories and 8 intents
2. Scores each prompt on clarity, specificity, constraints, reusability, and success
3. Extracts reusable templates from similar prompt patterns
4. Detects workflow sequences (Prompt Packs) from repeated session patterns
5. Generates prompt standards from your most effective prompts
6. Shows everything in a local dashboard with search, filters, and charts

Supported AI Tools

| Tool | Status | Data Location | |------|--------|--------------| | Claude Code | Working | ~/.claude/projects/ | | Cursor | Working | ~/.cursor/projects/ | | Codex CLI | Working | ~/.codex/sessions/ | | VS Code (Copilot) | Working | ~/Library/Application Support/Code/ | | Windsurf | Working | ~/Library/Application Support/Windsurf/ | | Zed | Working | ~/Library/Application Support/Zed/ | | Gemini CLI | Working | ~/.gemini/ | | Copilot CLI | Working | ~/.copilot/ | | Antigravity | Working | ~/Library/Application Support/Antigravity/ | | Goose | Working | ~/.goose/ | | Kiro | Working | ~/Library/Application Support/kiro/ | | OpenCode | Working | ~/.opencode/ | | Command Code | Working | ~/.commandcode/ |

13 adapters. Auto-detects which tools are installed on your machine.

Dashboard Pages

| Page | What It Shows | |------|--------------| | Overview | Optimization opportunities, weak prompts, top templates, activity trends | | Library | Searchable prompt library with quality/intent filters, weak/reusable badges | | Sessions | AI interaction sessions as timelines | | Projects | Per-project prompt patterns and stats | | Outcomes | Prompt effectiveness: file changes, follow-ups, abandonment risk | | Templates | Reusable prompt patterns with export, scoring, best examples | | Packs | Detected workflow sequences (e.g., "Bug Triage Pack", "Refactor & Test Pack") | | Standards | Best-practice prompt structures extracted from your most effective prompts | | Sources | Connected AI tools, scan status, discovery | | Settings | Database management, export, clear data |

Prompt Analysis

Every prompt gets analyzed with 6 quality scores:

• Clarity - How clear and unambiguous the prompt is
• Specificity - How specific vs vague
• Constraints - Whether output boundaries are defined
• Context Efficiency - How well context is provided without noise
• Ambiguity - Single clear request vs multiple unfocused ones
• Optimization - Overall quality (weighted average)

Plus refactor suggestions: strengths, weaknesses, anti-patterns, improved version, template version.

Tech Stack

| Layer | Technology | |-------|-----------| | Frontend | React 19, Vite, React Router, TanStack Query | | Styling | Tailwind CSS 4, Recharts, Framer Motion | | Backend | Express.js, TypeScript | | Database | SQLite (better-sqlite3), Drizzle ORM | | CLI | Node.js ESM |

Privacy

100% local. No cloud, no accounts, no API keys, no telemetry. Data stored in ~/.prompttrace/data/prompttrace.db. Open source under MIT.

Development

git clone https://github.com/ahmetseha/prompt-trace.git
cd prompt-trace
npm install
npm run dev

Starts Vite dev server + Express API with hot reload.

License

MIT