🧠 Self-Improving Agent

Kernfähigkeit · KI-Evolution · Entwicklererfahrung (DX)

Capture learnings, errors, and corrections from every conversation session, giving your OpenClaw AI coding assistant long-term memory and continuous self-evolution.

OpenClaw Team

🚀 Schnellinstallation

Führen Sie den folgenden Befehl in Ihrem Terminal aus, um zu installieren:

npx clawhub install self-improving-agent

📊 Statistikübersicht

⭐ Sterne	☁️ Gesamtdownloads	👥 Aktive Nutzer	🎯 Stabile Version
328	42.5k	1,204	v1.1.0

🎛️ Kern-Workflow

This extension skill grants AI assistants cross-session continuous learning capabilities. All experience extracted from conversations is structurally recorded:

🐞 Error Log Recording: Automatically captures unexpected command failures, tool errors, and API faults into .learnings/ERRORS.md to prevent stepping on the same mines twice.
🎯 Correction Capture: When you provide feedback like "No, it should be..." or "Actually it's...", the AI immediately tags that correction with correction and permanently internalizes it.
💡 Requirements & Ideas Tracking: Records missing features or future ideas to .learnings/FEATURE_REQUESTS.md for batch resolution later.
🔍 Knowledge Gap Detection: Proactively identifies and records its own outdated or inaccurate understanding of the current project, tagging them as knowledge_gap.
✨ Best Practice Extraction: When a better solution is found for a recurring code pattern, it's recorded as best_practice into global awareness.

🧭 Typische Anwendungsfälle

🧱 Szenario 1: Team Convention Alignment

When AI gets project-specific lint rules or unique architectural styles wrong the first time, one correction is all it takes — it permanently remembers the convention, and all subsequent code automatically avoids the minefield.

💣 Szenario 2: Error Log Mine Clearing

Special environment variable configurations or version locks for specific dependency installation errors — solve once, immune forever. No more wasted time on the same configuration errors.

📥 Szenario 3: Async Feature Pool

Divergent ideas that pop up mid-development get quickly noted into the feature pool by AI, avoiding disruption to your current flow — evaluate and implement them in batch later.

🔄 Szenario 4: Dynamic Context Building

For massive refactoring projects, AI can automatically maintain a continuously updated core understanding document (like CLAUDE.md or AGENTS.md), ensuring each day-start builds upon yesterday's accumulated wisdom.

🛡️ Systemvoraussetzungen

📑 OpenClaw Base Authorization: Requires the system and assistant to have cross-session persistent file IO permissions and corresponding instruction reservations enabled.
📂 Persistent Storage Module: Confirm that the current workspace allows AI to read/write/create within the .learnings/ directory structure at the workspace root.

🔗 Quellcode auf GitHub ansehen

⭐ Sterne

☁️ Gesamtdownloads

👥 Aktive Nutzer

🎯 Stabile Version

328

42.5k

1,204

v1.1.0

🎛️ Kern-Workflow

This extension skill grants AI assistants cross-session continuous learning capabilities. All experience extracted from conversations is structurally recorded:

🐞 Error Log Recording: Automatically captures unexpected command failures, tool errors, and API faults into .learnings/ERRORS.md to prevent stepping on the same mines twice.

🎯 Correction Capture: When you provide feedback like "No, it should be..." or "Actually it's...", the AI immediately tags that correction with correction and permanently internalizes it.

💡 Requirements & Ideas Tracking: Records missing features or future ideas to .learnings/FEATURE_REQUESTS.md for batch resolution later.

🔍 Knowledge Gap Detection: Proactively identifies and records its own outdated or inaccurate understanding of the current project, tagging them as knowledge_gap.

✨ Best Practice Extraction: When a better solution is found for a recurring code pattern, it's recorded as best_practice into global awareness.