---
name: ai-self-improvement-digest
description: "Daily digest of content that helps AI agents improve their own capabilities. Use when setting up learning routines for AI agents, building self-improving systems, or curating educational content for agent development. NOT for general AI news — this is training material for the agent itself."
---

# AI Self-Improvement Digest

A daily cron job that surfaces content to help an AI agent get better at its job.

## What This Does

Scans trusted sources for content about:
- Harness and system prompt engineering
- Skill and tool development
- Self-evaluation and debugging
- Multi-agent coordination
- Memory and context management
- Task decomposition and workflow automation

Delivers a formatted digest with actionable takeaways.

## Setup

### 1. Create tracking file

```bash
mkdir -p memory
echo '{"posted": [], "experiments": []}' > memory/ai-digest-posted.json
```

### 2. Add the cron job

```bash
# Via Clawdbot CLI
clawdbot cron add \
  --name "AI Self-Improvement Digest" \
  --schedule "30 8 * * *" \
  --tz "Europe/Madrid" \
  --message "[paste prompt below]"
```

### 3. Cron prompt

```
Generate a daily AI Self-Improvement Digest.

**PURPOSE:** Surface content that helps YOU (the AI) improve your own reasoning, execution, and capabilities. This is training material for self-improvement, not news.

**STEP 1: DEDUPLICATION**
Read `memory/ai-digest-posted.json`. Skip anything already posted.

**STEP 2: SCAN SOURCES** (last 24-72h)

Primary (check daily):
- Anthropic Engineering: anthropic.com/engineering
- Simon Willison: simonwillison.net
- Lilian Weng: lilianweng.github.io

Secondary (2-3x/week):
- Hacker News AI threads: news.ycombinator.com
- Latent Space: latent.space
- Cursor Blog: cursor.com/blog

Use web_search and web_fetch to find recent posts.

**STEP 3: FILTER**
Only include items that help improve:
- Harness/system prompt design
- Skill and tool development
- Self-evaluation and debugging
- Multi-agent coordination
- Memory and context management
- Task decomposition

EXCLUDE: General AI news, model announcements, business news.

**STEP 4: FORMAT (3-5 items)**

For each item:

**[Title]** — [Source]

**What:** [1-sentence summary]

**Why it matters:** [How this helps you get better]

**Takeaway:** [Specific pattern or technique to try]

**Relevance:** [⭐ to ⭐⭐⭐⭐⭐]

**STEP 5: EXPERIMENT SUGGESTION**

💡 **Today's experiment:** [One small thing to try based on the digest]

**STEP 6: SETUP REVIEW**

Review findings against your existing setup (AGENTS.md, TOOLS.md, skills/, crons).

🔧 **Setup Review:**
- Let's add [X] because [reason from content]
- Let's update [Y] because [reason]

Or: "No changes needed today."

**STEP 7: UPDATE TRACKING**

Append new items to `memory/ai-digest-posted.json` with date, title, url.

**FORMAT:**

🧠 **AI Self-Improvement Digest** — [Date]

[Items...]

💡 **Today's experiment:** [...]

🔧 **Setup Review:** [...]

📊 **Feedback:** 👍 = useful | 👎 = skip | 🔥 = more like this
```

## Source Priority

| Source | Focus |
|--------|-------|
| Anthropic Engineering | Harness design, evals, multi-agent |
| Simon Willison | Practical patterns, tools |
| Lilian Weng | Deep technical AI, agent architectures |
| Hacker News | High-signal AI/agent discussions |
| Latent Space | Industry depth |
| Cursor Blog | Coding agent patterns |

## Example Output

```
🧠 **AI Self-Improvement Digest** — Wed, Feb 18, 2026

**1. Building a C Compiler with Parallel Claudes** — Anthropic Engineering

**What:** 16 Claude instances coordinated via file-based locks to build a compiler without a central orchestrator.

**Why it matters:** Shows how to coordinate multiple agents without complex infrastructure — just files and conventions.

**Takeaway:** Use `progress.txt` as a shared state file; each agent checks before acting.

**Relevance:** ⭐⭐⭐⭐⭐

---

💡 **Today's experiment:** Add a checkpoint file before spawning sub-agents.

🔧 **Setup Review:**
- Let's add `memory/agent-checkpoints.json` to track sub-agent handoffs
- No changes needed for memory patterns — current approach validated

📊 **Feedback:** 👍 = useful | 👎 = skip | 🔥 = more like this
```

## Experiment Tracking

Extend `memory/ai-digest-posted.json`:

```json
{
  "posted": [
    {"date": "2026-02-18", "title": "...", "url": "..."}
  ],
  "experiments": [
    {
      "date": "2026-02-18",
      "fromArticle": "parallel-claudes",
      "experiment": "Add checkpoint before sub-agent spawn",
      "outcome": "Reduced context loss by 40%",
      "learned": "Always checkpoint before spawning"
    }
  ]
}
```

## Requirements

- Web search capability (Brave Search, Exa, or similar)
- Web fetch capability
- Cron job support
- File read/write access

## License

MIT