reflect

What This Skill Does

The reflect skill is a cornerstone of the OpenClaw architecture, enabling your agent to evolve from a static assistant into a continuously improving expert. It functions as a meta-cognition layer that audits conversation history to identify user corrections, successful patterns, and architectural insights. By applying the philosophy of 'Correct once, never again,' the reflect skill parses high-confidence signals—such as explicit 'never' or 'always' instructions—and permanently encodes these learnings into the agent's definition files or dedicated skill manifests. This eliminates the 'forgetting' cycle common in conversational AI, ensuring that your preferred coding styles, architectural choices, and behavioral nuances persist across long-term development sessions.

Installation

To begin improving your agent's knowledge base, initialize the skill via the OpenClaw command-line interface:

clawhub install openclaw/skills/skills/stevengonsalvez/reflect-learn

Once installed, you can toggle automatic analysis by running reflect on. The agent will then monitor the interaction context during every turn, providing periodic summaries of learned behaviors and pending updates for your approval.

Use Cases

• Project Onboarding: Teach an agent your specific library preferences or strict project-wide coding standards after the initial setup phase.
• Complex Debugging: When you and your agent spend significant time resolving a misleading error, use reflect to ensure the root cause and the specific resolution path are indexed for future recurrence.
• Behavioral Tuning: If your agent has a tendency to be overly verbose or misses specific formatting requirements, a quick correction followed by a reflect cycle ensures the instruction is permanently adopted.
• Orchestration: Use it to capture lessons learned when coordinating between multiple backend and frontend specialists, mapping those lessons to the shared CLAUDE.md context.

Example Prompts

1. "We finally fixed that API timeout issue. Reflect on this session and extract the fix into a permanent guideline for our backend developer role."
2. "Reflect on the last three hours of coding. I want you to capture our new pattern for modularizing React components and update the frontend developer definitions."
3. "Reflect status: show me what learnings you have staged for integration and summarize the confidence levels for each."

Tips & Limitations

• Review Before Commit: While reflect is powerful, always review proposals via reflect review before pushing changes to your production agent definitions to prevent logic regressions.
• Confidence Thresholds: The system prioritizes HIGH confidence signals. If you are training the agent on a subtle nuance, explicitly state it clearly in the chat to ensure it is picked up by the parser.
• Quality Control: Not every interaction is a learning moment. Use reflect strategically after finishing a task to keep the signal-to-noise ratio in your configuration files optimal.