A deterministic execution framework that forces AI agents to operate exclusively through external memory state files for maximum reliability.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install memory-driven-execution-loop
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 memory-driven-execution-loop 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 Memory-Driven Execution Loop?
The Memory-Driven Execution Loop is a high-precision protocol designed for Openclaw Skills to ensure AI agents remain grounded and deterministic. By offloading internal state to a series of markdown-based memory files, this skill eliminates hallucination and drift, forcing the agent to validate its environment and goals before every single action. It is particularly effective for complex, multi-step engineering tasks where auditability and strict adherence to rules are paramount.
This approach shifts the agent from a reactive model to a proactive, state-aware system. By utilizing a specific set of memory files—including rules, goals, and lessons—the agent maintains a persistent context that survives session restarts and provides a clear trail of reasoning for every execution step taken.
Memory-Driven Execution Loop Use Cases
- Managing complex software refactoring tasks that require multi-step state tracking.
- Executing timed research or development sessions with strict wall-clock requirements.
- Building self-correcting agent loops that convert execution failures into structured, cross-domain lessons.
- Standardizing AI behavior across different Openclaw Skills through a shared memory protocol.
How Memory-Driven Execution Loop Works
- The agent initializes by reading the rules.md file to establish the operational constraints.
- All secondary memory files (goal, plan, progress, notes, and lessons) are ingested to build a complete situational awareness.
- The system performs a preflight check to validate exactly one clear next action based on the current plan.
- Reasoning for the intended action is documented in notes.md before execution occurs.
- The agent executes exactly one step of the plan to prevent cascading errors.
- Real-time updates are written to progress.md and notes.md to reflect the new state.
- If an error occurs, it is processed through a strict Lesson Quality Gate and recorded in lessons.md.
Memory-Driven Execution Loop Setup
To implement this within your environment, ensure the following markdown files exist in your working directory:
touch rules.md goal.md plan.md progress.md notes.md lessons.md
Configure your agent system prompt to recognize these files as the primary source of truth for all Openclaw Skills execution loops.
Memory-Driven Execution Loop Data Schema & Taxonomy
| File | Purpose | Key Metadata Requirements |
|---|---|---|
| rules.md | Constraints | Global operational boundaries and logic gates. |
| goal.md | Objective | The high-level desired end-state. |
| plan.md | Strategy | Sequential steps required to reach the goal. |
| progress.md | Tracking | Current status, completed steps, and time logs. |
| notes.md | Reasoning | Thinking process and justification for each step. |
| lessons.md | Optimization | Failed attempts, triggers, and boundary conditions. |
Memory-Driven Execution Loop Advanced Features
- Strict Timed Task Rule: Forces adherence to wall-clock duration with mandatory start/end/elapsed proof for time-boxed activities.
- Universal Lesson Quality Gate: Ensures every lesson includes a trigger, action, metric, and at least two cross-field examples to ensure maximum utility.
- Deterministic One-Step Execution: Prevents the agent from over-reaching or skipping validation steps during complex workflows.
- Externalized State Management: Makes the agent's internal logic fully transparent and editable by human collaborators.
SKILL.md
Loading
Related Openclaw Skills
Model Pricing Calculator
A specialized tool for fetching, unifying, and calculating standardized pricing ratios across multiple AI model providers and API platforms.
360 Power Saver
An intelligent power management skill that provides one-click energy optimization, deep battery health analysis, and real-time consumption diagnostics.
Bili-Trending Skill
A specialized Openclaw tool that extracts the top 20 trending videos from Bilibili including titles, summaries, and direct links.
Accuranker for Openclaw
A powerful SEO integration for Openclaw Skills that automates keyword tracking, domain analysis, and competitor monitoring through the Membrane framework.
Mysteel Bid & Supply Query
A specialized tool for steel industry professionals to query real-time supply-demand data and historical bidding information.
CN Express Tracker
A high-performance logistics tracking skill that connects to the Kuaidi100 API to monitor domestic Chinese and international shipments.
Featured*
