| name | self-improving-agent |
| version | 1.0.0 |
| description | Structured learning and error logging system for agents that get better over time. Use when implementing continuous improvement loops, error correction, or agent learning systems. Covers LRN/ERR/FEAT entry format, .learnings directory management, priority/status tracking, metadata tagging, and promotion workflow to integrate learnings into CLAUDE.md/AGENTS.md. Enables agents to detect failure patterns and autonomously promote fixes. |
| source | https://lobehub.com/skills/openclaw-skills-self-improving-agent-1-0-0 (4.86/5 rating) |
| changelog | Initial release — structured agent learning system |
Self-Improving Agent
Structured learning and error logging system that enables agents to continuously improve through experience.
Learning Entry Format
LRN (Learnings)
New knowledge gained during operation:
LRN [PRIORITY:HIGH|MEDIUM|LOW] [STATUS:ACTIVE|PROMOTED|ARCHIVED]
Context: What situation triggered this learning
Observation: What was discovered
Application: How to apply this going forward
Tags: [category, domain, pattern]
ERR (Errors)
Mistakes and failures with corrective actions:
ERR [PRIORITY:HIGH|MEDIUM|LOW] [STATUS:OPEN|FIXED|PROMOTED]
Context: What was being attempted
Root Cause: Why it failed
Fix: What corrects this
Prevention: How to avoid recurrence
Tags: [error-type, domain, severity]
FEAT (Feature Requests)
Improvements identified during operation:
FEAT [PRIORITY:HIGH|MEDIUM|LOW] [STATUS:PROPOSED|IMPLEMENTED|PROMOTED]
Motivation: Why this would help
Description: What it should do
Implementation: How to build it
Tags: [feature-type, domain, effort]
Directory Structure
.learnings/
├── entries/
│ ├── 2026-05-12-lrn-react-hydration.md
│ ├── 2026-05-12-err-ssr-random-values.md
│ └── 2026-05-12-feat-auto-dedup.md
├── index.md # Searchable index of all entries
└── promoted/ # Learnings promoted to system prompts
Priority Tracking
| Priority | Response Time | Promotion Threshold |
|---|
| HIGH | Immediate fix needed | After 1 confirmed occurrence |
| MEDIUM | Fix in next session | After 2 confirmed occurrences |
| LOW | Fix when convenient | After 3+ confirmed occurrences |
Detection Triggers
The agent should create learning entries when:
- A task fails or produces incorrect output
- A user corrects the agent's approach
- A pattern is noticed across multiple interactions
- A workaround is discovered for a recurring issue
- A better approach is found than what was originally used
Promotion Workflow
- Detect — Error or learning pattern identified
- Log — Create LRN/ERR/FEAT entry in
.learnings/entries/
- Verify — Confirm the learning is valid (not a one-off)
- Promote — Integrate into AGENTS.md or relevant SKILL.md
- Archive — Move original entry to
.learnings/promoted/
Activation Triggers
- Building agent systems that learn from experience
- Implementing error correction loops
- Creating agent memory that improves over time
- Designing continuous improvement workflows for AI agents
