scribe

# Scribe

An autonomous "stenographer" agent that watches your OpenClaw session history and extracts what matters into persistent, structured memory files — so your agent remembers across sessions without relying on context window compaction.

## Install & Setup

**1. Copy the skill to your workspace:**
```bash
cp -r skills/public/scribe ~/.openclaw/workspace/skills/public/scribe
```

**2. Register the nightly cron job (one command):**
```bash
python3 skills/public/scribe/scripts/setup-cron.py
```

That's it. Scribe will run every night at 23:30 and write `memory/YYYY-MM-DD.md` to your workspace.

**3. Run manually anytime:**
```bash
python3 skills/public/scribe/scripts/scribe.py
```

## How It Works

1. Scans today's session JSONL files from `~/.openclaw/agents/main/sessions/`
2. Filters out heartbeats, system messages, and noise
3. Sends user messages to an LLM (via OpenRouter, reads your existing OpenClaw API key)
4. Extracts: decisions, preferences, framework sentences, project updates, todos
5. Writes structured output to `{workspace}/memory/YYYY-MM-DD.md`

## Output Format

```markdown
# YYYY-MM-DD Memory (Scribe)

## 🔑 Decisions Made
## 💡 Preferences & Rules
## 🗣️ Framework Sentences
## 📦 Project Updates
## ✅ Todos / Follow-ups
```

## Configuration

Environment variables (all optional — defaults work out of the box):

| Variable | Default | Description |
|---|---|---|
| `SCRIBE_SESSION_DIR` | `~/.openclaw/agents/main/sessions` | Session JSONL location |
| `SCRIBE_WORKSPACE` | `~/.openclaw/workspace` | Where memory files are written |
| `SCRIBE_DAYS` | `1` | Days back to scan |
| `SCRIBE_MODEL` | `anthropic/claude-haiku-4-5` | LLM model (OpenRouter ID) |
| `SCRIBE_APPEND_LONGTERM` | `false` | Also append to MEMORY.md |
| `OPENROUTER_API_KEY` | *(from OpenClaw config)* | Override API key |

## References

- **Cron setup details**: `references/cron-setup.md` — manual config, launchd, timezone changes
- **Signal classification**: `references/signal-guide.md`...

# scribe — OpenClaw Memory Skill

> An autonomous stenographer for your AI agent. Extracts decisions, preferences, and key insights from your daily sessions and writes them to persistent memory files.

## The Problem

AI agents forget everything between sessions. Context window compaction is lossy and passive. You end up re-explaining yourself constantly.

## The Solution

Scribe runs every night as an isolated cron job, scans your OpenClaw session logs, and extracts what actually matters — structured, searchable, durable.

```markdown
# 2026-03-04 Memory (Scribe)

## 🔑 Decisions Made
- Chose Plan B (cron + skill) for the stenographer architecture

## 🗣️ Framework Sentences
- "一步一步来" — prefers incremental execution
- "做不到就换个方法" — zero tolerance for blocking

## 📦 Project Updates
- scribe skill: scripts complete, publishing to ClawdHub
```

## Install

```bash
# 1. Install via ClawdHub (once published)
clawdhub install scribe

# 2. Register nightly cron
python3 skills/public/scribe/scripts/setup-cron.py

# 3. Run manually
python3 skills/public/scribe/scripts/scribe.py
```

## Requirements

- [OpenClaw](https://openclaw.ai) with an active gateway
- OpenRouter API key (reads from your existing OpenClaw config automatically)

## Files

```
scribe/
├── SKILL.md                    # Skill definition + onboarding
├── scripts/
│   ├── scribe.py               # Core extraction script
│   └── setup-cron.py           # One-command cron registration
└── references/
    ├── cron-setup.md           # Cron config details
    └── signal-guide.md         # How signals are classified
```

## License

MIT