A dedicated system for capturing execution lessons, managing corrections, and maintaining memory hygiene for AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install retrospective-agent
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 retrospective-agent 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 Retrospective Agent?
The Retrospective Agent is a specialized framework within the ecosystem of Openclaw Skills designed to facilitate continuous improvement for AI coding agents. Unlike standard memory systems that store raw facts and events, this skill focuses exclusively on execution-memory hygiene. It provides a structured methodology for logging what went well, identifying where tools failed, and refining future workflows without overstepping guardrails or rewriting agent identity. This approach ensures that Openclaw Skills remain predictable and efficient by promoting proven success patterns over time through an auditable and manual-approval-first process.
Retrospective Agent Use Cases
- Capturing reusable corrections when a user provides specific feedback on agent performance.
- Conducting post-task reviews to identify tool failure patterns or success metrics after complex operations.
- Generating weekly retrospective reports to provide a birds-eye view of recurring wins and misses.
- Standardizing project or domain-specific execution lessons to be shared across a workspace.
- Logging repeated tool failures to suggest better alternative workflows in future Openclaw Skills interactions.
How Retrospective Agent Works
- Detects specific triggers such as the completion of a multi-step task or an explicit user request for a review.
- Separates factual data (dates, people, facts) into existing memory files while routing execution lessons to the skill specific ops folder.
- Logs entries using a specific scope hierarchy ranging from project-specific to global execution lessons.
- Applies a conservative promotion model (Observed, Repeated, Candidate) to ensure changes are only recommended after repeated evidence.
- Generates compact, operational reports that recommend improvements without autonomously altering core agent configurations or identity files.
Retrospective Agent Setup
To initialize the Retrospective Agent within your environment, ensure the following directory structure is created in your workspace:
# Create the necessary directories for the skill
mkdir -p workspace/skills/retrospective-agent/
mkdir -p workspace/ops/retrospective-agent/{weekly,domains,projects,templates}
# Initialize core correction log
touch workspace/ops/retrospective-agent/corrections.md
Retrospective Agent Data Schema & Taxonomy
The skill organizes data across two main directories to maintain a clean separation of concerns within Openclaw Skills.
| Directory | Content Type |
|---|---|
workspace/skills/ |
Contains the core skill logic and reference documentation for workflow boundaries. |
workspace/ops/ |
Stores operational data including corrections, weekly reviews, and project-specific lessons. |
Key Operational Files:
corrections.md: A central log of captured corrections and classified patterns.weekly/: A dedicated folder for recurring performance reviews and candidate rule updates.domains/: Execution lessons scoped to specific technical domains (e.g., frontend, devops).projects/: Scoped lessons unique to individual project repositories and coding standards.
Retrospective Agent Advanced Features
- Automated Promotion Model: Implements a strict threshold (1 occurrence: observed, 2: repeated, 3: candidate rule) to suggest workflow changes safely.
- Scoped Hierarchy Logic: Intelligently classifies lessons into Project, Domain, or Global execution scopes to prevent logic leakage between unrelated tasks.
- Template-Driven Reporting: Utilizes standardized markdown templates for post-task retrospectives and weekly summaries to keep logs consistent.
- Non-Intrusive Guardrails: Explicitly prevents the skill from modifying sensitive identity files like SOUL.md or USER.md, maintaining the integrity of Openclaw Skills.
- Evidence-Based Recommendations: Requires documented repeated patterns before suggesting a confirmation rule, ensuring that agent behavior remains stable and predictable.
SKILL.md
Loading
Related Openclaw Skills
PDF Manipulation and Processing
A comprehensive toolkit for automating PDF creation, extraction, and modification tasks within your AI agent environment.
ComfyUI Image Generation
Connect your AI agent to a local ComfyUI instance to generate images programmatically using natural language prompts.
Binance Earning Assistant
A specialized monitoring tool that aggregates and summarizes real-time earning opportunities, airdrops, and promotional events from Binance.
Meta Ads Analytics
A specialized analytical skill for diagnosing Meta Ads performance and generating structured operational reports via the Meta MCP.
Spec Coding
Spec Coding is a specification-driven development workflow that automates the creation and reverse-engineering of technical documentation for features, APIs, and UI components.
AI Automation Consulting
A specialized consulting service designed to automate repetitive business tasks and save labor costs using advanced AI frameworks.
Featured*
