Agente Conhecimento

What This Skill Does

The Agente Conhecimento (Knowledge Agent) is a powerful meta-cognitive tool for OpenClaw that enables your AI agents to engage in continuous self-improvement. Rather than forgetting past interactions, this skill provides a structured framework for documenting, categorizing, and applying lessons learned from errors, user corrections, and technical roadblocks. By integrating this skill, your agent effectively creates a persistent feedback loop, turning every failure or knowledge gap into a permanent upgrade in its operational efficiency. It forces the agent to monitor its performance, track system-level errors, and consolidate project-specific knowledge into centralized documentation files that govern future interactions.

Installation

To install this skill, you must ensure your OpenClaw environment is ready to handle persistent markdown logs. The recommended installation method is via the command-line interface:

1. Use the ClawdHub installer: clawdhub install openclaw/skills/skills/pedrohenrique202525/agente-conhecimento
2. If installing manually, clone the repository into your skills directory: git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
3. Initialize the workspace structure by executing: mkdir -p ~/.openclaw/workspace/.learnings. This folder serves as the central repository for your LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md files.

Use Cases

• Error Resolution: When a script, API call, or CLI command fails, the agent automatically logs the failure context to ERRORS.md to prevent future cycles of the same mistake.
• Knowledge Refinement: When a user provides a correction (e.g., "Actually, the correct format is X"), the agent uses the 'correction' category to update its internal best practices.
• Feature Tracking: When you request a capability that isn't currently supported, the agent logs it to FEATURE_REQUESTS.md for future development roadmap consideration.
• Refactoring: After identifying a more efficient way to perform a recurring task, the agent updates best practices, ensuring your workflow becomes faster and more reliable over time.

Example Prompts

1. "I noticed you keep failing on the build step. Check the error log, analyze why it keeps happening, and update the best practices so you don't repeat this in the next iteration."
2. "Actually, you should prioritize the secondary database instead of the primary one when performing backups. Please record this as a correction in your knowledge base."
3. "Review the current LEARNINGS.md file and summarize the top three improvements we have made to our development workflow this week."

Tips & Limitations

• Consistency is Key: The agent's intelligence is strictly proportional to how often you ask it to review its logs. Schedule periodic 'Knowledge Review' sessions.
• Maintain Focus: Don't let your LEARNINGS.md become cluttered. Use the 'promote to AGENTS.md' workflow mentioned in the instructions to keep the information actionable.
• Privacy Warning: Because this tool logs errors and corrections to disk, ensure that you are not logging sensitive API keys, passwords, or personal user data into the markdown files within the .learnings directory. The skill is designed for architectural and workflow knowledge, not secrets management.