planning-with-files

What This Skill Does

Planning-with-files brings persistent memory to your AI agent workflows, effectively treating your filesystem as an external hard drive for the agent's context window. By maintaining task_plan.md, findings.md, and progress.md in your project root, the agent can resume complex multi-step tasks after interruptions or long context cycles. It follows the Manus-style planning pattern, ensuring that critical data, research findings, and error logs are stored on disk rather than relying solely on the volatile short-term memory of the LLM. It includes built-in session recovery scripts to sync state after a /clear command.

Installation

To integrate this skill into your environment, run the following command in your terminal: clawhub install openclaw/skills/skills/othmanadi/planning-with-files Ensure your environment has Python installed, as the session-catchup utility requires it to bridge the gap between sessions.

Use Cases

• Complex Software Development: Manage multi-feature branches or architectural refactors where context spans multiple days.
• Deep Research Projects: Compile data from browser searches or documentation reading into a persistent summary that doesn't disappear when the context window reaches its limit.
• Multi-step Automation: Execute long-running data processing tasks where error handling and progress tracking are essential for reliability.
• Iterative Debugging: Track bug discovery, reproduction steps, and failed attempts to build a persistent log of what didn't work.

Example Prompts

1. "I need to refactor the authentication module across the entire codebase. Start by setting up a task_plan.md and scanning the existing directory structure."
2. "We hit an error with the API integration. Please update the findings.md with the latest error trace and suggest a modification to the task plan."
3. "I am resuming our research project. Run the session-catchup script, check the current task_plan.md, and let's continue from the last incomplete phase."

Tips & Limitations

• The 2-Action Rule: Always save findings to your markdown files after every two tool calls (browser, search, or file read) to avoid losing information.
• Proactive Maintenance: Treat the markdown files as the source of truth; if the agent seems to wander, ask it to "Read the task_plan.md" to realign its objectives.
• Error Tracking: Be diligent about filling out the 'Errors Encountered' table in your plan file—this is the most effective way to prevent the agent from repeating the same failed attempts in future sessions.