Planning with Files
VerifiedClaude Code skill implementing Manus-style persistent markdown planning.
Install
Claude Code
Add to .claude/skills/ About This Skill
Planning with Files is a skill that brings Manus-style persistent planning to Claude Code by maintaining structured markdown files as living task plans. Instead of keeping plans only in the conversation context where they get lost as the thread grows, this skill instructs the agent to create and continuously update a markdown planning document on disk.
When activated, the agent begins every complex task by writing a plan.md file (or similar) that breaks the work into numbered steps with checkboxes. As each step is completed, the agent updates the file to mark it done, add notes about decisions made, and adjust remaining steps based on what was learned. This creates an auditable trail of progress that survives context window limits and can be reviewed after the session ends.
The planning file follows a consistent structure: an objective statement at the top, a numbered task list with status indicators (pending, in-progress, done, blocked), notes on key decisions or findings, and a summary section updated upon completion. The agent re-reads the plan file before each major action to stay aligned with the overall goal, which is particularly valuable in long sessions where earlier context may be truncated.
This approach solves a fundamental problem with AI coding agents: they tend to lose track of multi-step goals in long conversations, skip steps, or repeat work. By externalizing the plan to a file, the agent has a reliable single source of truth that persists regardless of context window pressure.
The skill is especially effective for refactoring projects, multi-file feature implementations, migration tasks, and any work that spans more than a few steps. It pairs well with other skills that benefit from structured task tracking, and the resulting plan files serve as useful documentation of what was done and why.
Use Cases
- Managing a multi-file refactoring project with dozens of sequential changes across the codebase
- Tracking progress on a complex migration task that spans multiple sessions
- Coordinating a feature implementation that requires changes to API, database, and frontend layers
- Creating an auditable record of AI agent decisions and progress for team review
Pros & Cons
Pros
- + Prevents the agent from losing track of goals in long conversations
- + Creates persistent documentation of plans and decisions on disk
- + Works naturally with Claude Code's file read/write capabilities
Cons
- - Adds overhead for simple, single-step tasks that don't need formal planning
- - Plan files need manual cleanup if you don't want them in your repo
- - Only available for Claude Code, not transferable to other agent platforms
Stay Updated on Agent Skills
Get weekly curated skills + safety alerts
每周精选 Skills + 安全预警