When should I use the Planning with Files skill?

It is recommended for any task requiring more than five tool calls, multi-step project organization, or intensive research tasks.

What is Manus-style file-based planning?

It is an agentic pattern where the AI uses the filesystem as 'Disk' memory to store plans, findings, and logs, ensuring it doesn't lose track of goals when the chat context (RAM) is cleared or filled.

Can I recover my progress if the session restarts?

Yes, the skill includes a session-catchup script that reads current planning files and git diffs to restore the AI's understanding of the project's current state.

How does this skill help with long Claude Code sessions?

It provides a standardized way to save progress to markdown files, allowing you to use the /clear command to refresh the context window without losing the project roadmap or research data.

What are the core files used by this skill?

The skill initializes and maintains three key files: task_plan.md (for phases and decisions), findings.md (for research and discoveries), and progress.md (for session logs and results).

Planning with Files

Name: Planning with Files
Author: Shawn-Guo-CN

byShawn-Guo-CN

0•

生産性とワークフロー

Implements persistent file-based planning to manage and track progress on complex, multi-step Claude Code projects.

This skill transforms how Claude handles complex tasks by implementing a 'working memory on disk' pattern inspired by high-end agents. By automatically creating and maintaining specialized markdown files—task_plan.md, findings.md, and progress.md—it enables the AI to track state across long sessions, recover from cleared contexts, and maintain a clear roadmap. It is essential for large-scale engineering tasks, deep research, or any workflow requiring significant tool orchestration where volatile context window limits might otherwise cause lost progress or repetitive errors.

主な機能

01Structured task decomposition using a centralized, phase-based planning file.

02Persistent storage of research findings and key discoveries to bypass context limits.

03Strategic 'Read Before Decide' workflows to improve decision accuracy in complex environments.

04Mandatory error logging protocols to ensure the AI never repeats failed actions.

050 GitHub stars

06Automated session recovery and context syncing via specialized catchup scripts.

ユースケース

01Executing complex multi-step feature development requiring more than 5 tool operations.

02Managing large-scale codebase refactors or migrations that span multiple sessions.

03Conducting deep-dive technical research and documentation across numerous web sources.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add shawn-guo-cn/agentfoundry-plugin-marketplace planning-with-files

For use in Claude.ai and ChatGPT

Download Skill