Self-Improving Agent

Captures learnings, errors, and corrections to enable continuous improvement. This skill creates a feedback loop where the agent learns from mistakes, user corrections, and discoveries.

When to Use

1. A command or operation fails unexpectedly → Log to ERRORS.md
2. User corrects Claude ("No, that's wrong...", "Actually...") → Log to LEARNINGS.md
3. User requests a capability that doesn't exist → Log to FEATURE_REQUESTS.md
4. An external API or tool fails → Log to ERRORS.md
5. Claude realizes knowledge is outdated/incorrect → Log to LEARNINGS.md
6. A better approach is discovered → Log to LEARNINGS.md

Log Files

All logs are stored in .learnings/:

Workflow

1. Capture a Learning

When you learn something new or are corrected:

1. Read the appropriate log file
2. Append the new learning with timestamp
3. Save the file

2. Review Before Major Tasks

Before starting significant work:

1. Read LEARNINGS.md for relevant past discoveries
2. Read ERRORS.md to avoid repeating mistakes
3. Apply lessons learned to current task

3. Periodic Consolidation

Weekly or monthly:

1. Review all logs
2. Extract high-value learnings to MEMORY.md
3. Clear or archive processed entries

Log Format

LEARNINGS.md

## 2026-02-20 - Docker Compose Network Issue

**Context:** User needed to connect containers
**Discovery:** Use `docker network create` + `external: true` in compose
**Correction:** Previously suggested deprecated `--link` flag
**Apply to:** All Docker networking tasks

ERRORS.md

## 2026-02-20 - Git Clone Permission Denied

**Command:** `git clone [email protected]:...`
**Error:** Permission denied (publickey)
**Cause:** SSH key not added to GitHub
**Fix:** Use HTTPS instead, or add SSH key
**Prevention:** Check auth before suggesting SSH

FEATURE_REQUESTS.md

## 2026-02-20 - Notion Database Query

**Request:** Query Notion database with filters
**Status:** Not implemented
**Priority:** High
**Notes:** Would need Notion integration skill

Integration with MEMORY.md

High-value learnings should be promoted to MEMORY.md:

• Critical fixes that prevent repeat mistakes
• User preferences discovered through corrections
• Tool-specific quirks and workarounds
• Security or safety lessons

Example Session

User: "Actually, use gh auth login not gh auth status"

Agent:

1. Read ~/.openclaw/workspace/skills/self-improving-agent/.learnings/LEARNINGS.md
2. Append:
   ## 2026-02-20 - GitHub CLI Auth Correction
   
   **User correction:** Use `gh auth login` not `gh auth status`
   **Context:** When authenticating GitHub CLI
   **Lesson:** `gh auth login` is the interactive flow; `gh auth status` only checks
   **Apply to:** All GitHub CLI auth tasks
3. Write file
4. Respond: "Noted — I'll use `gh auth login` for GitHub CLI authentication going forward."

Best Practices

• Be specific — Include full context, not just the correction
• Tag with dates — Helps identify stale learnings
• Link to tasks — Reference what you were working on
• Review regularly — Don't let learnings go to waste
• Promote to MEMORY.md — Important ones become long-term memory

Related Skills

• agent-autonomy-kit — Proactive behavior patterns
• triple-memory-skill — Persistent memory system
• flowmind — Task and goal management