compound-engineering-2（复合工程 2）

Compound Engineering

Make your AI agent learn automatically. Extract learnings from sessions, update memory files, and compound knowledge over time.

The idea: Your agent reviews its own work, extracts patterns and lessons, and updates its instructions. Tomorrow's agent is smarter than today's.

Quick Start

# Review last 24 hours and update memory
npx compound-engineering review

# Create hourly memory snapshot
npx compound-engineering snapshot

# Set up automated nightly review (cron)
npx compound-engineering setup-cron

How It Works

The Compound Loop

┌─────────────────────────────────────────┐
│           DAILY WORK                    │
│  Sessions, chats, tasks, decisions      │
└────────────────┬────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────┐
│        NIGHTLY REVIEW (10:30 PM)        │
│  • Scan all sessions from last 24h      │
│  • Extract learnings and patterns       │
│  • Update MEMORY.md and AGENTS.md       │
│  • Commit and push changes              │
└────────────────┬────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────┐
│        NEXT DAY                         │
│  Agent reads updated instructions       │
│  Benefits from yesterday's learnings    │
└─────────────────────────────────────────┘

What Gets Extracted

• Patterns: Recurring approaches that worked
• Gotchas: Things that failed or caused issues
• Preferences: User preferences discovered
• Decisions: Key decisions and their reasoning
• TODOs: Unfinished items to remember

Clawdbot Integration

Automatic Hourly Memory

Add to your HEARTBEAT.md:

# Hourly Memory Snapshot
Every hour, append a brief summary to memory/YYYY-MM-DD.md:
- What was accomplished
- Key decisions made
- Anything to remember

Or use cron:

# Add to clawdbot config or crontab
0 * * * * clawdbot cron run compound-hourly

Nightly Review Job

Add this cron job to Clawdbot:

{
  "id": "compound-nightly",
  "schedule": "30 22 * * *",
  "text": "Review all sessions from the last 24 hours. For each session, extract: 1) Key learnings and patterns, 2) Mistakes or gotchas to avoid, 3) User preferences discovered, 4) Unfinished items. Update MEMORY.md with a summary. Update memory/YYYY-MM-DD.md with details. Commit changes to git."
}

Manual Review Command

When you want to trigger a review manually:

Review the last 24 hours of work. Extract:

1. **Patterns that worked** - approaches to repeat
2. **Gotchas encountered** - things to avoid
3. **Preferences learned** - user likes/dislikes
4. **Key decisions** - and their reasoning
5. **Open items** - unfinished work

Update:
- MEMORY.md with significant long-term learnings
- memory/YYYY-MM-DD.md with today's details
- AGENTS.md if workflow changes needed

Commit changes with message "compound: daily review YYYY-MM-DD"

Memory File Structure

MEMORY.md (Long-term)

# Long-Term Memory

## Patterns That Work
- When doing X, always Y first
- User prefers Z approach for...

## Gotchas to Avoid
- Don't do X without checking Y
- API Z has rate limit of...

## User Preferences
- Prefers concise responses
- Timezone: PST
- ...

## Project Context
- Main repo at /path/to/project
- Deploy process is...

memory/YYYY-MM-DD.md (Daily)

# 2026-01-28 (Tuesday)

## Sessions
- 09:00 - Built security audit tool
- 14:00 - Published 40 skills to MoltHub

## Decisions
- Chose to batch publish in parallel (5 sub-agents)
- Security tool covers 6 check categories

## Learnings
- ClawdHub publish can timeout, retry with new version
- npm publish hangs sometimes, may need to retry

## Open Items
- [ ] Finish remaining MoltHub uploads
- [ ] Set up analytics tracker

Hourly Snapshots

For more granular memory, create hourly snapshots:

# Creates memory/YYYY-MM-DD-HH.md every hour
*/60 * * * * echo "## $(date +%H):00 Snapshot" >> ~/clawd/memory/$(date +%Y-%m-%d).md

Or have the agent do it via heartbeat by checking time and appending to daily file.

The Compound Effect

Week 1: Agent knows basics Week 2: Agent remembers your preferences Week 4: Agent anticipates your needs Month 2: Agent is an expert in your workflow

Knowledge compounds. Every session makes future sessions better.

Setup Scripts

Nightly Review (launchd - macOS)

<!-- ~/Library/LaunchAgents/com.clawdbot.compound-review.plist -->
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "...">
<plist version="1.0">
<dict>
  <key>Label</key>
  <string>com.clawdbot.compound-review</string>
  <key>ProgramArguments</key>
  <array>
    <string>/opt/homebrew/bin/clawdbot</string>
    <string>cron</string>
    <string>run</string>
    <string>compound-nightly</string>
  </array>
  <key>StartCalendarInterval</key>
  <dict>
    <key>Hour</key>
    <integer>22</integer>
    <key>Minute</key>
    <integer>30</integer>
  </dict>
</dict>
</plist>

Hourly Memory (crontab)

# Add with: crontab -e
0 * * * * /opt/homebrew/bin/clawdbot cron run compound-hourly 2>&1 >> ~/clawd/logs/compound.log

Best Practices

1. Review before sleep - Let the nightly job run when you're done for the day
2. Don't over-extract - Focus on significant learnings, not noise
3. Prune regularly - Remove outdated info from MEMORY.md monthly
4. Git everything - Memory files should be version controlled
5. Trust the compound - Effects are subtle at first, dramatic over time

Built by LXGIC Studios - @lxgicstudios