criticism-self-criticism

What This Skill Does

The criticism-self-criticism skill is a foundational meta-cognitive tool for the OpenClaw AI agent, designed to implement the dialectical process of structured review and continuous improvement. It functions by enforcing a rigorous framework for identifying, analyzing, and correcting errors within any workflow. Rooted in the principles of structured self-reflection, this skill helps the agent move beyond mere task completion into a cycle of 'cleaning' its own process, much like washing one's face or cleaning a room to prevent the accumulation of intellectual dust. By utilizing this skill, you enable the agent to maintain high quality standards, process feedback without defensiveness, and foster long-term performance growth.

Installation

To integrate this skill into your environment, run the following command in your terminal: clawhub install openclaw/skills/skills/arg0/criticism-self-criticism

Use Cases

• Post-Delivery Review: After finalizing a code module or report, invoke this skill to ensure all quality gates were met.
• Feedback Processing: When a human supervisor or another agent provides negative input, use this to parse the critique and generate a remediation plan.
• Error Recurrence: If a similar technical error appears multiple times, use this to conduct a root cause analysis rather than applying temporary patches.
• Project Milestones: Use at the end of a sprint or development phase to hold a retrospective and refine the working methodology.
• Developmental Coaching: Deploy this skill when the agent needs to objectively evaluate its own internal decision-making processes.

Example Prompts

1. "I've just finished the refactoring of the authentication module. Please run a self-criticism check to identify any shortcuts I might have taken."
2. "The user feedback indicates the generated summary was too verbose. Process this criticism, identify why the error occurred, and explain how to improve for the next task."
3. "We keep seeing timeout errors in the data extraction flow. Let's run a retrospective using the criticism-self-criticism framework to find the root cause."

Tips & Limitations

• Be Specific: The tool is most effective when you provide the specific context or output being reviewed. Generic requests yield generic feedback.
• Avoid Subjectivity: While the tool is designed to be introspective, it relies on objective facts. Ensure the underlying data (logs, code, previous prompts) is provided.
• Non-Punitive Intent: The goal is systemic improvement. Frame your prompts around 'how can we improve' rather than 'who is to blame'.
• Mechanism, Not Just Event: Integrate this into your standard operational workflow rather than waiting for a critical failure. Regular 'cleansing' prevents minor oversights from becoming systemic failures.