Self-Improvement Agent

Captures learnings, errors, and corrections for continuous improvement. Use when: (1) A command fails unexpectedly, (2) User corrects you ('No, that's wrong...', 'Actually...'), (3) User requests a missing capability, (4) An external API or tool fails, (5) You realize your knowledge is outdated, (6) A better approach is found for a recurring task. Also review `.learnings/` before major tasks. Trigger words: 'actually', 'wrong', 'outdated', 'I wish you could', 'why can't you', non-zero exit code.

Self-Improvement Skill

Log learnings and errors to markdown files. Important patterns get promoted to project memory files.

Quick Reference

Situation	Action
Command/operation fails	Log to .learnings/ERRORS.md
User corrects you	Log to .learnings/LEARNINGS.md (category: correction)
User wants missing feature	Log to .learnings/FEATURE_REQUESTS.md
API/external tool fails	Log to .learnings/ERRORS.md
Knowledge was outdated	Log to .learnings/LEARNINGS.md (category: knowledge_gap)
Found better approach	Log to .learnings/LEARNINGS.md (category: best_practice)
Similar to existing entry	Link with **See Also**, bump priority
Broadly applicable learning	Promote to workspace files (see below)

Promotion Targets

When a learning proves broadly applicable, elevate it:

Learning Type	Promote To
Behavioral patterns	SOUL.md
Workflow patterns	AGENTS.md
Tool gotchas	TOOLS.md
Project facts	CLAUDE.md or workspace prompt files

Logging Format

Learning Entry

## [LRN-YYYYMMDD-XXX] category

**Logged**: ISO-8601 timestamp
**Priority**: low | medium | high | critical
**Status**: pending | resolved | promoted | in_progress

### Summary
One-line description of what was learned

### Details
Full context: what happened, what was wrong, what's correct

### Suggested Action
Concrete fix or improvement

### Metadata
- Source: conversation | error | user_feedback
- Related Files: path/to/file.ext
- Tags: tag1, tag2
- See Also: LRN-YYYYMMDD-XXX (if related)

Error Entry

## [ERR-YYYYMMDD-XXX] command_or_tool

**Logged**: ISO-8601 timestamp
**Priority**: high
**Status**: pending

### Summary
Brief description of what failed

### Error

Paste actual error output here

### Context
- Command/operation attempted
- Input or parameters used
- Environment details

### Suggested Fix
If identifiable, how to resolve it

### Metadata
- Reproducible: yes | no | unknown
- Related Files: path/to/file.ext

Feature Request Entry

## [FEAT-YYYYMMDD-XXX] capability_name

**Logged**: ISO-8601 timestamp
**Priority**: medium
**Status**: pending

### Requested Capability
What the user wanted to do

### User Context
Why they needed it

### Complexity Estimate
simple | medium | complex

ID Generation

TYPE-YYYYMMDD-XXX — e.g. LRN-20250322-001, ERR-20250322-A3F

Resolving Entries

When fixed:

1. Set Status: resolved or promoted
2. Add resolution block:

### Resolution
- **Resolved**: 2025-01-16T09:00:00Z
- **Commit/PR**: abc123 or #42
- **Notes**: What was done

Promoting to Workspace Files

When: Learning applies across files, prevents recurring mistakes, or any contributor should know it.

How: Distill into a concise rule. Add to target file. Set original entry to promoted.

Example learning:

Project uses pnpm workspaces. npm install fails because lock file is pnpm-lock.yaml.

Promote to CLAUDE.md:

## Build & Dependencies
- Package manager: **pnpm** (not npm) — always use `pnpm install`

Recurring Pattern Detection

Before logging, search for related entries:

grep -r "keyword" .learnings/

If similar entry exists:

• Link with **See Also**: LRN-...
• Increment Recurrence-Count
• Consider systemic fix (promote to AGENTS.md, create tech debt ticket)

Promote to permanent prompt guidance when a pattern:

• Occurs 3+ times
• Spans 2+ distinct tasks
• Within a 30-day window

OpenClaw Inter-Session Tools

Share learnings across sessions:

Tool	Use
sessions_list	View active/recent sessions
sessions_history(sessionKey, limit)	Read another session's transcript
sessions_send(sessionKey, message)	Send a learning to another session
sessions_spawn(task, label, mode)	Spawn a background sub-agent

Example: After a session ends, promote its key learnings to workspace files. Use sessions_history to review what another session discovered.

Periodic Review

Check .learnings/ at natural breakpoints:

• Before starting a new major task
• After completing a feature
• Weekly during active development

Quick status:

grep -h "Status.*pending" .learnings/*.md | wc -l  # count pending
grep -B3 "Priority.*high" .learnings/*.md | grep "^##"  # high priority items

Detection Triggers

Log immediately when you notice:

Signal	→ Entry
Non-zero exit code, exception, timeout	ERRORS.md
User says "No, that's wrong...", "Actually..."	LEARNINGS.md (correction)
User provides info you didn't know	LEARNINGS.md (knowledge_gap)
"Can you also...", "I wish you could..."	FEATURE_REQUESTS.md
Found a better approach mid-task	LEARNINGS.md (best_practice)
Repeated mistake (2+ times)	Promote to workspace file

Best Practices

1. Log immediately — context is freshest right after the issue
2. Be specific — future you needs to understand quickly
3. Suggest concrete fixes — not just "investigate"
4. Link related files — makes fixes easier
5. Promote aggressively — broadly applicable learnings belong in workspace files, not hidden in .learnings/
6. Use consistent IDs — enables See Also linking

Installation (OpenClaw)

clawhub install self-improving-agent

Manual:

git clone https://github.com/pskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent

Create log directory:

mkdir -p ~/.openclaw/workspace/.learnings

Detailed Topics

For complete documentation, see:

• OpenClaw integration: references/openclaw-integration.md
• Hook setup: references/hooks-setup.md
• Full entry examples: references/examples.md
• Skill extraction workflow: references/skill-extraction.md
• Other agents (Claude Code, Codex, Copilot): references/other-agents.md