Planning with Files is an agent skill that provides persistent file-based planning for AI coding agents and long-running tasks.
The tool solves the problem of context loss in AI agents by maintaining three markdown files on disk—task_plan.md, findings.md, and progress.md—that survive context window resets, /clear commands, crashes, and compaction. Rather than relying on prompts the agent might follow, it uses lifecycle hooks that fire every turn to inject planning context back into the agent's working memory. This approach keeps the plan persistent while the agent's context window dies, allowing recovery to read only project files without requiring replay of session records unless explicitly requested through a catchup mode.
The tool suits developers working with long-running autonomous agents across multiple coding platforms. It installs as an agent skill compatible with over 60 agents including Claude, Cursor, Codex, and GitHub Copilot, available through npm, the Claude Code plugin marketplace, or npx. The three-file pattern separates task planning, findings, and progress tracking, creating a deterministic completion gate that prevents context rot from derailing multi-turn agent work.
The project shows active maintenance with blind A/B testing demonstrating measurable improvements in agent task completion. Development includes comprehensive evaluation documentation and a full installation guide, indicating sustained attention to both functionality and user onboarding.
