Neura Market
Planning With Files logo

Planning With Files

Free

Claude Code skill implementing Manus-style persistent markdown planning — the workflow pattern behind the $2B acquisition.

FreeFree tier
Type
Open Source
Visit Website GitHub (19.0k stars)

About Planning With Files

Planning with Files is a persistent file-based planning skill for AI coding agents. It keeps three markdown files (task_plan.md, findings.md, progress.md) on disk and re-injects them every turn, so the plan survives context loss, /clear commands, crashes, and compaction. It implements a Manus-style working memory with an optional completion gate. The skill supports over 60 agents including Claude Code, Codex, Cursor, Kiro, OpenCode, and many others through the Agent Skills standard. In benchmark tests, fresh sessions with the files on disk resumed in 5.0 turns on average versus 13.3 turns for a raw agent (internal v1, author-run).

Key Features

Persistent file-based planning via task_plan.md, findings.md, and progress.md
Per-turn re-injection of plan data against context rot
Survives context window loss, /clear, crashes, and compaction
Opt-in deterministic completion gate
Manus-style working memory on disk
Supports 60+ agents including Claude Code, Codex, Cursor, Kiro, OpenCode via Agent Skills standard
Session recovery benchmark: 5.0 turns vs 13.3 turns without the skill

Pros & Cons

Pros
  • Eliminates loss of context after /clear or crashes
  • Works with a wide range of AI coding agents (60+)
  • Uses simple markdown files for transparency and debuggability
  • Lightweight open-source implementation (MIT license)
  • Improves task completion rate (96.7% pass rate with skill in internal tests)
Cons
  • Requires initial setup and configuration for each agent
  • Relies on file system access, which may not suit all environments
  • The benchmark data is internally run by the author; external validation may vary

Best For

Long-running AI coding agent tasks that require persistent stateRecovery after context window reset or agent session /clearManaging complex multi-phase development projects with AI assistantsEnsuring continuity across multiple agent invocations

FAQ

What is Planning with Files?
It is a persistent file-based planning skill for AI coding agents. It stores a task plan, findings, and progress in markdown files on disk and re-injects them every turn, so the plan survives context loss, /clear, crashes, and compaction.
Which agents are supported?
The skill supports over 60 agents including Claude Code, Codex, Cursor, Kiro, OpenCode, and many others via the Agent Skills standard. See the AGENTS.md file in the repository for the full list.
How does it handle context loss?
The skill keeps task_plan.md, findings.md, and progress.md on disk. When a new session starts, the plan data is re-injected from those files, allowing the agent to resume where it left off.
Is there a performance benchmark?
In internal benchmarks (author-run), a fresh session with the files on disk resumed in 5.0 turns on average, compared to 13.3 turns for a raw agent without the skill.
What is a completion gate?
The completion gate is an opt-in feature that ensures tasks are marked complete only when explicitly determined, preventing the agent from incorrectly marking unfinished tasks as done.