planning-with-files provides persistent, file-based markdown planning and completion gating to help AI coding agents survive context loss and handle long-running tasks.
planning-with-files is an open-source persistent file-based planning system designed for AI coding agents and long-running tasks. It works across over 60 agents (including Claude Code, Codex, and Cursor) by storing durable Markdown files—specifically task_plan.md, findings.md, and progress.md—directly on disk, making the agent's memory and plan crash-proof against context loss or command-line clears. Its recent update introduces opt-in autonomous and gated modes featuring a deterministic completion gate that prevents the agent from finishing until all planned tasks are fully resolved, mimicking Manus-style workflow persistence.
AI agents are notoriously forgetful during context clearing or long tasks, and planning-with-files resolves this by forcing them to use the local filesystem as a durable working memory.
* Durable Memory: Storing state in Markdown files on disk acts as a persistent memory bank that survives terminal restarts, crashes, and chat clears.
* Multi-Agent Compatibility: Extensive cross-agent support via specialized folders makes it compatible with Claude Code, Codex CLI, Cursor, Kiro, Gemini, and others.
* Gatekeeper Mechanism: The new completion gating ensures that autonomous agents cannot declare a task finished prematurely without passing deterministic validation checks.
DISCOVERED
1h ago
2026-07-05
PUBLISHED
1h ago
2026-07-05
RELEVANCE