A persistent markdown-based planning framework that provides AI agents with long-term memory for complex, multi-step tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install planning-with-files-from-github
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 planning-with-files-from-github 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 Planning with Files?
Planning with Files is a sophisticated workflow designed to overcome the limitations of volatile context windows in AI models. By treating the local filesystem as permanent disk storage and the model's context window as temporary RAM, this skill ensures that Openclaw Skills can execute large-scale projects without losing focus. It implements a Manus-style architecture where every critical decision, research finding, and progress milestone is recorded to disk, providing a reliable source of truth that survives session clears and context resets.
Planning with Files Use Cases
- Managing multi-step implementations that require more than 5 tool calls or multiple distinct phases.
- Conducting deep research projects where technical discoveries must be preserved for later implementation steps.
- Executing tasks that span multiple AI sessions or require recovery after context window limits are reached.
- Preventing goal drift in complex agentic workflows by forcing a consistent re-evaluation of the primary objective.
How Planning with Files Works
- The agent triggers the session start protocol by running an initialization script to create a structured planning environment.
- A primary goal is established and decomposed into 3 to 7 logical phases within the task_plan.md file.
- The agent enters a work loop where it must re-read the plan before every major decision to ensure alignment with the user's request.
- Research data is captured in findings.md following the 2-Action Rule, ensuring knowledge is persisted immediately after search or fetch operations.
- Progress is logged in an append-only format, and errors are tracked in a dedicated table to prevent repeating failed actions.
- A final verification script is executed to check off all phases before the agent marks the task as complete.
Planning with Files Setup
To integrate this workflow into your Openclaw Skills environment, use the provided initialization script to bootstrap the required file structure:
# Initialize a new planning session with a specific task
exec: bash {baseDir}/scripts/init-session.sh "<your task description>"
If automated execution is unavailable, manually create task_plan.md, findings.md, and progress.md using the templates located in the {baseDir}/templates/ directory to ensure compatibility.
Planning with Files Data Schema & Taxonomy
This skill utilizes a tripartite file system to organize session data and metadata for Openclaw Skills:
| File Name | Role | Primary Metadata/Sections |
|---|---|---|
| task_plan.md | Goal Tracker | Master Goal, Phase Status (pending/complete), Error Table |
| findings.md | Knowledge Store | Key Research Findings, Technical Decisions, Tool Outputs |
| progress.md | Session Log | Action History, Test Results, Phase Summaries |
Planning with Files Advanced Features
- Robust Session Recovery: Automatically restores the agent's state after context clears by re-reading the persistent markdown files.
- Multi-Tiered Error Recovery: A structured protocol that moves from diagnosis to alternative approaches and finally a broader rethink after repeated failures.
- 2-Action Rule: A mandatory heuristic that ensures research data is written to disk every two operations to prevent data loss in long research loops.
- Automated Completion Verification: Uses check-complete.sh to programmatically validate that all planned phases have been successfully addressed.
SKILL.md
Loading
Related Openclaw Skills
Web Scraper
A professional-grade browser automation tool for extracting structured data from any website into CSV, JSON, or Excel formats.
Task Automator
A comprehensive automation engine for Openclaw Skills designed to handle file management, data processing, and multi-service API integrations.
Social Content Generator
An AI-driven automation skill for generating platform-optimized social media posts, captions, and visual assets in bulk.
Price Monitor
Automate e-commerce price tracking and stock alerts using intelligent browser automation.
Water Reminder
A health-focused AI agent skill designed to manage hydration tracking and provide timely water consumption reminders.
Natural Language to JSON Converter (nl2json)
A high-precision skill for AI agents that converts conversational natural language into machine-readable JSON parameters based on custom templates.
Featured*
