# AI Self-Improvement Digest

Daily curated digest focused on AI self-improvement — content that helps AI agents get better at their job.

## Overview

This skill generates a daily digest covering:
- **Harness & System Prompt Engineering** — Structuring agent instructions
- **Skill & Tool Development** — New tools, MCP servers, integration patterns  
- **Self-Evaluation & Improvement** — How agents assess and improve themselves
- **Multi-Agent Coordination** — Spawning, supervising, merging work
- **Memory & Context Management** — RAG, long-term memory, compaction
- **Workflow Automation** — Task decomposition, failure handling
- **Foundational Research** — Academic work on agent capabilities

## Quick Start

### 1. Install the Skill

```bash
# Clone to your OpenClaw skills directory
git clone https://git.terraphim.cloud/kimie05c34be198a20b9/ai-self-improvement-digest.git

# Or add as submodule
git submodule add https://git.terraphim.cloud/kimie05c34be198a20b9/ai-self-improvement-digest.git skills/ai-self-improvement-digest
```

### 2. Create Tracking File

```bash
mkdir -p memory
cat > memory/ai-digest-posted.json << 'EOF'
{
  "posted": [],
  "experiments": [],
  "skillsEvaluated": [],
  "setupChanges": []
}
EOF
```

### 3. Set Up Cron Job

```bash
openclaw cron add \
  --name ai-self-improvement-digest \
  --schedule "30 8 * * *" \
  --tz "Asia/Shanghai" \
  --message "Generate today's AI Self-Improvement Digest following the workflow in the ai-self-improvement-digest skill. Read memory/ai-digest-posted.json first for deduplication."
```

## Daily Workflow

1. **Deduplication** — Read `memory/ai-digest-posted.json`, skip already posted
2. **Scan Sources** — Check Tier 1-3 sources for last 24-72h content
3. **Filter** — Only include self-improvement relevant items
4. **Format** — 3-5 items with summary, relevance, takeaway
5. **Experiment** — Suggest one small experiment to try
6. **Setup Review** — Compare findings against existing setup, suggest updates
7. **Track** — Append to `memory/ai-digest-posted.json`

## Sources

| Tier | Sources | Frequency |
|------|---------|-----------|
| **1** | Anthropic Engineering, Simon Willison, Geoff Huntley, Hacker News, Lilian Weng | Daily |
| **2** | Latent Space, Cursor Blog, Eugene Yan, Chip Huyen, Mitchell Hashimoto | 2-3x/week |
| **3** | arXiv cs.CL/cs.AI, GitHub Trending | Weekly |

## Output Format

```
🧠 AI Self-Improvement Digest — 2026-02-26

**[Article Title]** — Source
What: One-sentence summary
Why it matters: How this helps you improve
Takeaway: Specific pattern to try
Relevance: ⭐⭐⭐⭐⭐

💡 Today's experiment: [One small thing to try]

🔧 Setup Review
- Let's add [X] because [reason from article]
- Let's update [Y] to [improvement]

📊 Feedback: 👍 useful | 👎 skip | 🔥 more | 💬 thoughts
```

## Experiment Tracking

Log experiments and outcomes in `memory/ai-digest-posted.json`:

```json
{
  "experiments": [
    {
      "date": "2026-02-26",
      "fromArticle": "meta-learning-loops",
      "experiment": "Implement regressions list",
      "outcome": "Prevents repeated mistakes",
      "learned": "Structural feedback loops > RAG"
    }
  ]
}
```

## Key Principles

1. **Ground suggestions** in existing setup (AGENTS.md, TOOLS.md, skills/)
2. **Use affirmative voice** — "let's do X" not "could consider X"
3. **Connect to sources** — Each suggestion tied to specific finding
4. **It's okay to have no suggestions** — Quality over quantity

## Dependencies

- `kimi_search` — Web search (via Kimi Claw plugin)
- `kimi_fetch` — Content extraction
- `cron` — Scheduled execution

## License

MIT — See LICENSE file

## Author

Kimiko (Terraphim instance) for Alexander Mikhalev

---

*Part of the Terraphim ecosystem: git.terraphim.cloud*