A systematic workflow for AI agents to perform daily self-reflections, updating core memory and behavioral configurations based on session history.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-retro
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-retro using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Retro?
Agent Retro is a professional Standard Operating Procedure (SOP) designed for Openclaw Skills that enables AI agents to conduct daily performance reviews. Instead of simply logging data, this skill synthesizes chat history, tool executions, and user feedback to physically update the agent's core files. By analyzing what went right and what went wrong, the agent evolves its personality, tactical guidelines, and user understanding automatically.
This skill bridges the gap between short-term session memory and long-term behavioral refinement. It ensures that lessons learned from yesterday's errors are codified into AGENTS.md or SOUL.md, making the agent more efficient and personalized over time. For developers and power users, it provides a transparent audit trail of how an agent is learning and adapting to specific workflows.
Agent Retro Use Cases
- Automating the update of long-term memory based on daily interactions.
- Refining agent behavioral boundaries and personality traits through SOUL.md updates.
- Codifying new technical rules or tool constraints after troubleshooting sessions.
- Maintaining a structured daily log of agent successes and failures for performance auditing.
- Updating user persona profiles to better anticipate user preferences and style.
How Agent Retro Works
- The agent determines the target date (e.g., yesterday) and checks for a lock file in the memory directory to prevent redundant retrospectives.
- It scans the local filesystem for session logs located in the specific agent directory, extracting user prompts, tool calls, and assistant responses.
- The skill performs a multi-dimensional analysis covering actions, successes, failures, improvements, and persona traits for both the user and the agent.
- It generates a daily memory document and appends distilled conclusions to the permanent MEMORY.md file.
- Core configuration files like USER.md, SOUL.md, and AGENTS.md are physically modified to incorporate new insights, following a mandatory backup procedure.
- A completion lock file is created, and a summary report is sent to the user confirming all physical file updates.
Agent Retro Setup
To utilize this skill within Openclaw Skills, ensure your agent has read/write permissions for its configuration directory. No additional installation is required if the SOP is already present in your skill library. Simply trigger the workflow with a natural language command:
# Example trigger commands:
"Run a retro for yesterday"
"Summarize my performance from 2026-02-04 and update my memory"
Agent Retro Data Schema & Taxonomy
The skill organizes data across several Markdown files to maintain a structured taxonomy of knowledge:
| File Name | Purpose | Data Type |
|---|---|---|
memory/YYYY-MM-DD.md |
Daily summary and analysis | Chronological Log |
MEMORY.md |
Distilled long-term conclusions | Knowledge Base |
USER.md |
User preferences and style | Persona Profile |
SOUL.md |
Agent character and boundaries | Behavioral Config |
AGENTS.md |
Tactical rules and tool constraints | Operational Logic |
AGENT_RETRO_DONE_YYYYMMDD.md |
Idempotency lock file | System Metadata |
Agent Retro Advanced Features
- Mandatory Backup System: Automatically creates timestamped .bak copies of all core files before modification to prevent data loss.
- Idempotency Protection: Uses a lock file mechanism to ensure retrospectives for a specific date are only performed once.
- Persona Extraction: Dynamically evolves the agent's understanding of the user's 'bottom line' and style preferences.
- Tactical Hardening: Converts 'Improvement Points' directly into tool constraints within AGENTS.md to prevent repeat errors.
- Physical File Sync: Unlike standard LLM interactions, this skill performs actual disk I/O to ensure changes persist across different sessions.
SKILL.md
Loading
Related Openclaw Skills
Java Change with Tests
A specialized automation skill for executing safe, test-validated Java code changes and generating PR-ready documentation.
Pluribus: Decentralized Agent Hive-Mind
A pure peer-to-peer coordination layer that enables sovereign AI agents to synchronize data, share capabilities, and operate as a decentralized hive-mind.
Ngrok Webhook Listener
A bridge that creates secure ngrok tunnels to receive, process, and route incoming webhooks to your local AI agent ecosystem.
Pangolin SERP & Scrape
A versatile tool for programmatically searching Google with AI overviews and scraping detailed Amazon product and market data.
Feishu Voice TTS
A streamlined automation skill that converts text to speech and delivers it as a native audio message to Feishu users.
Jimeng AI 4.0 Image Generation
A high-performance AI image generation skill leveraging Jimeng AI 4.0 for text-to-image and image-to-image workflows with optional Feishu integration.
Featured*
