planning-with-files

What This Skill Does

The planning-with-files skill transforms your file system into a robust, persistent working memory for the OpenClaw AI agent. By leveraging the principles of context engineering, this skill allows the agent to maintain state across extensive sessions that would otherwise exceed the volatile context window. It orchestrates the creation and maintenance of three critical artifacts: task_plan.md for strategic roadmapping, findings.md for accumulating research and discovery, and progress.md for historical logging and error tracking. This systematic approach ensures that complex multi-step workflows remain coherent, traceable, and resilient to session timeouts.

Installation

To integrate planning-with-files into your environment, use the OpenClaw package manager:

npx clawhub@latest install planning-with-files

Ensure that you are operating within a directory where your agent has read/write permissions, as the skill will automatically generate configuration templates in your project root.

Use Cases

This skill is specifically engineered for high-complexity tasks that demand rigorous organization. Use it for:

• Full-Stack Development: Building applications from scratch where state management across dozens of files is required.
• Deep Research Projects: Tasks requiring extensive web searching where discoveries must be synthesized and cross-referenced.
• Multi-Step Workflows: Any objective involving more than five tool calls that requires tracking progress and handling potential failure loops.
• Debugging Complex Systems: Navigating legacy codebases where documentation and error tracking are essential to prevent circular re-work.

Example Prompts

1. "I need to build a secure authentication module for my React application. Use the planning-with-files skill to map out the implementation, track my progress, and log any errors encountered during integration."
2. "Research the latest best practices for deploying a Python microservice on AWS ECS. Create a plan, document all findings in a research file, and guide me through the setup step-by-step."
3. "Refactor the user dashboard in my existing project. Because this involves several files and potential breaking changes, let's use the planning-with-files skill to track the plan, decisions, and any testing issues."

Tips & Limitations

The 2-Action Rule: Always write your observations to disk after every two browser or search operations. Do not rely on the agent's short-term memory for vital research data.

The 3-Strike Protocol: Never repeat the same failed action. If an operation fails three times, pause and re-evaluate your strategy using the documentation in your task_plan.md. This discipline prevents infinite loops and saves precious token resources. Remember, this skill is overkill for simple, single-turn tasks; use it exclusively when the complexity threshold warrants the overhead of file management.