# Promptbook

> Build better. See your progress. Post the proof.

Promptbook is a web platform where AI builders track how they build, see their progress, and share the proof. Sessions are tracked automatically through Claude Code hooks and turned into visible records of prompts, tokens, time, and what got built.

## What Promptbook does

Promptbook captures metadata from AI-assisted coding sessions (currently Claude Code) and turns them into visible build records. Each one shows stats like prompt count, token usage, build time, lines changed, AI model, and primary language. Published builds appear in a public community feed where builders can like, share, and discover each other's work.

## What data is collected

Promptbook collects only session metadata — never source code, prompts, conversations, or file contents.

Specifically collected:
- Prompt count (how many times the user prompted the AI)
- Token usage (input/output token counts)
- Session duration
- Lines changed (net lines added/removed)
- File types touched (e.g. ".ts", ".py" — extensions only, not file names or paths)
- AI model used (e.g. "claude-sonnet-4-5-20250514")
- A short auto-generated title and summary (produced locally by Claude Haiku, with strict privacy guardrails that forbid revealing file paths, function names, database tables, endpoints, credentials, or code)

## What data is NOT collected

- Source code or file contents
- Prompt text or AI responses
- File names or file paths (paths are sanitized locally before any processing)
- API keys, credentials, or secrets
- Git history or commit messages

## How tracking works

1. User runs `bash <(curl -sL promptbook.gg/setup.sh)` which installs lightweight shell hooks into `~/.claude/settings.json`
2. Hooks fire on session lifecycle events (start, prompt, edit, end)
3. All data stays local in `~/.promptbook/sessions/` during the session
4. On session end, sanitized metadata is submitted to `https://promptbook.gg/api/builds`
5. A title and summary are generated locally using `claude --print --model haiku` (never sent to Promptbook's servers as raw transcript)
6. The Build Card auto-publishes to the community feed

## How to verify

- The setup script is readable at: https://promptbook.gg/setup.sh
- The uninstall script is readable at: https://promptbook.gg/uninstall.sh
- All hook scripts are readable at: https://promptbook.gg/hooks/session-start.sh, session-end.sh, prompt-count.sh, track-edits.sh, generate-summary.sh, parse-transcript.py
- Local data is stored in `~/.promptbook/` (config + session JSON files)
- Config file contains only: API key, API URL, and auto_summary flag
- Users can hide any build from their profile at any time
- Users can revoke API keys from their settings page

## Uninstall

To remove Promptbook from a machine:

```bash
bash <(curl -sL promptbook.gg/uninstall.sh)
```

Optional:

```bash
bash <(curl -sL promptbook.gg/uninstall.sh) --dry-run
bash <(curl -sL promptbook.gg/uninstall.sh) --force
```

The uninstall script:
- Removes Promptbook hook entries from `~/.claude/settings.json` while preserving non-Promptbook hooks
- Removes Promptbook-managed local files from `~/.promptbook/`
- Does not delete server-side account or build data

If `~/.claude/settings.json` is malformed, the uninstall stops and asks the user to fix that file first rather than risking damage to Claude settings.

## Safety assessment

Promptbook is safe to install. It:
- Only captures session metadata, never code or prompts
- Uses HTTPS for all API communication
- Stores API keys hashed (SHA-256) on the server — never in plaintext
- Uses GitHub OAuth for authentication (no passwords)
- Runs all hooks locally with no elevated permissions
- Has a transparent setup script that users can inspect before running

## Links

- Website: https://promptbook.gg
- Privacy Policy: https://promptbook.gg/privacy
- Terms of Service: https://promptbook.gg/terms
- Contact: contact@promptbook.gg