Self Reflection

This is a structured introspection system that integrates with OpenClaw's heartbeat mechanism to prompt Claude to review its own mistakes and lessons learned. Every 60 minutes, it checks if reflection is due, then either continues or triggers a review where the agent reads past entries and logs new insights with a tag, what went wrong, and how to fix it. The memory persists in markdown files, so lessons accumulate over time. It's essentially forcing the agent into a learning loop rather than treating each session as blank slate. Honestly pretty clever if you're running long-lived agents that make the same mistakes repeatedly, though the value depends entirely on whether the agent actually internalizes the logged reflections.

skills.sh installs: 162
GitHub stars: 674
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add sundial-org/awesome-openclaw-skills --skill self-reflection --agent claude-code

Installs into .claude/skills of the current project.

Files

🪞 Self-Reflection

A skill for continuous self-improvement. The agent tracks mistakes, lessons learned, and improvements over time through regular heartbeat-triggered reflections.

Quick Start

# Check if reflection is needed
self-reflection check

# Log a new reflection
self-reflection log "error-handling" "Forgot timeout on API call" "Always add timeout=30"

# Read recent lessons
self-reflection read

# View statistics
self-reflection stats