planning-with-files-zh

Organizes complex multi-step tasks using persistent Markdown planning files on disk.

Updated May 28, 2026
One-click install
npx skills add https://github.com/changfengpro/agent-skills --skill planning-with-files-zh-changfengpro
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: planning-with-files-zh
Source: https://github.com/changfengpro/agent-skills/tree/main/skills/planning-with-files-zh
Command: npx skills add https://github.com/changfengpro/agent-skills --skill planning-with-files-zh-changfengpro

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires orjson, and includes scripts (resource) components.

What problem does it solve? Long-running AI-assisted tasks lose context when sessions are compacted or cleared, causing repeated work, forgotten goals, and lost research. This Skill persists plans, findings, and progress as Markdown files in your project directory so work survives context resets and session interruptions. ## Core Features & Use Cases - Three-File Planning System: Creates task_plan.md (phases and decisions), findings.md (research and discoveries), and progress.md (session logs and test results) in your project root. - Session Recovery: A session-catchup script detects unsynced context from previous Claude Code or Codex sessions and suggests how to restore state after /clear or compaction. - Lifecycle Hooks: UserPromptSubmit, PreToolUse, PostToolUse, Stop, and PreCompact hooks re-inject plan state, remind you to update progress, and verify plan integrity via SHA-256 attestation to block prompt injection through tampered plan files. - Use Case: When asked to plan and execute a multi-phase research or build project requiring many tool calls, the Skill initializes the three files, tracks phase status (pending → in_progress → complete), logs errors, and resumes cleanly after interruptions. ## Quick Start Ask the AI to create a task plan for your multi-step project using the planning-with-files system and initialize the planning files in the current project directory.

Frequently Asked Questions about planning-with-files-zh

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I keep an AI coding assistant on track during long multi-step tasks?

Use file-based planning: create task_plan.md, findings.md, and progress.md in your project root so goals and state persist outside the context window. This Skill automates creation, updating, and re-reading of these files across the session.

How to recover AI session context after /clear or compaction?

Run the session-catchup.py script, which parses previous Claude Code or Codex session transcripts, finds the last planning file update, and prints unsynced messages. Then run git diff --stat and update the planning files before continuing.

Does this planning skill work on Windows?

Yes. It ships both Bash and PowerShell versions of the init-session and check-complete scripts, and the Stop hook detects the platform to invoke the appropriate one. The Python catchup script also handles Windows and Git Bash path formats.

When should I not use file-based task planning?

Skip it for simple questions, single-file edits, and quick queries where planning overhead exceeds the benefit. It is designed for tasks with three or more steps, research work, or anything spanning many tool calls.

How does the skill prevent prompt injection through plan files?

Hooks treat plan file contents as data, not instructions, and support SHA-256 attestation of task_plan.md. If the file hash differs from the recorded attestation, injection is blocked and the user is asked to re-approve the plan.