SkillsApr 4, 2026·2 min read

Planning with Files — Manus-Style Persistent Planning Skill

Claude Code skill implementing persistent markdown planning with 96.7% benchmark pass rate. Uses a 3-file pattern (task_plan.md, findings.md, progress.md) to survive context resets.

TO
TokRepo精选 · Community
Quick Use

Use it first, then decide how deep to go

This block should tell both the user and the agent what to copy, install, and apply first.

Install the planning skill globally: ```bash npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g ``` Or manually: 1. Copy the SKILL.md content below 2. Save to `.claude/skills/planning-with-files.md` in your project 3. Claude Code loads it automatically on next conversation ---
Intro
Planning with Files is a Claude Code skill that implements Manus-style persistent markdown planning, with 18,000+ GitHub stars and a 96.7% benchmark pass rate (29/30 assertions vs 6.7% without). It uses a 3-file pattern — `task_plan.md` for tracking phases, `findings.md` for storing research, and `progress.md` for session logs — treating the filesystem as persistent disk while the context window is volatile RAM. The skill survived 3/3 blind A/B tests and supports 16+ platforms. Best for developers working on complex, multi-session projects where context continuity matters. Works with: Claude Code, Cursor, Codex CLI, Gemini CLI, Kiro, Copilot, Mastra Code. Setup time: under 1 minute. ---
## The 3-File Pattern The core concept is simple but powerful: ``` Context Window = RAM (volatile) Filesystem = Disk (persistent) → Anything important gets written to disk. ``` ### task_plan.md — Track phases and progress ```markdown # Task Plan ## Phase 1: Research - [x] Investigate existing solutions - [x] Document findings - [ ] Choose approach ## Phase 2: Implementation - [ ] Build core feature - [ ] Add tests ``` ### findings.md — Store research and discoveries ```markdown # Findings ## API Options - Option A: REST API (simpler, documented) - Option B: GraphQL (flexible, newer) → Decision: Option A — better docs, team familiarity ``` ### progress.md — Session log and test results ```markdown # Progress ## Session 2026-04-04 - Completed Phase 1 research - Found 3 API options, chose REST - Next: Start Phase 2 implementation ``` ### Key Features - **Session Recovery**: When context resets, the skill reads planning files to restore full project context - **2-Action Rule**: After every 2 significant actions, update planning files - **3-Strike Error Protocol**: If the same error occurs 3 times, document it in findings.md and try a different approach - **Security Boundaries**: Planning files never contain secrets, API keys, or credentials ### Benchmark Results | Metric | With Skill | Without Skill | |--------|-----------|--------------| | Assertion Pass Rate | 96.7% (29/30) | 6.7% (2/30) | | Blind A/B Wins | 3/3 | 0/3 | | Context Recovery | Full | None | ### FAQ **Q: What is Planning with Files?** A: A Claude Code skill that implements persistent markdown planning using a 3-file pattern, achieving 96.7% benchmark pass rate for complex multi-session projects. **Q: Is Planning with Files free?** A: Yes, fully free and open source under the MIT license. **Q: How do I install Planning with Files?** A: Run `npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g` for global installation. ---
🙏

Source & Thanks

> Created by [Ahmad Othman Ammar Adi](https://github.com/OthmanAdi). Licensed under MIT. > > [planning-with-files](https://github.com/OthmanAdi/planning-with-files) — ⭐ 18,000+ Thank you for bringing structured planning persistence to AI coding workflows.

Discussion

Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.

Related Assets

Awesome AI Agents 2026 — 340+ Tools Directory

The most comprehensive directory of AI agents, frameworks, and tools in 2026. Covers 340+ resources across 20+ categories from coding agents to voice AI, updated monthly.

TokRepo精选

Best of MCP Servers — 450 Ranked MCP Servers

A weekly-updated ranked directory of 450 MCP servers across 34 categories with quality scores based on GitHub stars, contributors, and activity. Find the best MCP server for any use case.

TokRepo精选

VoltAgent — TypeScript AI Agent Framework

Open-source TypeScript framework for building AI agents with built-in Memory, RAG, Guardrails, MCP, Voice, and Workflow support. Includes LLM observability console for debugging.

TokRepo精选
◈Home🔍Search👤Me