planning-with-files

Track multi-step task progress by using persistent markdown files as external working memory. You maintain a task_plan.md for phases, notes.md for findings, and a deliverable file, reading them before decisions and updating after actions to preserve goals and errors.

File-based memory is a pattern that treats the filesystem as persistent storage for task plans, research notes, and deliverables. It enforces read-before-decide and update-after-act workflows, allowing agents to recover from errors and catch up on session context across many tool calls.

Set up persistent notes by creating task_plan.md, findings.md, and a deliverable file in your working directory. Use the provided helper scripts to initialize sessions, report progress, check completion, and catch up unsynced context for feature development and bug fixes.

Yes, file-based planning works for research tasks by maintaining findings.md for discoveries and logging errors within the markdown files. This allows robust recovery and periodic reads, ensuring progress and decisions are preserved across multiple tool calls.

Use markdown files for session management when complex projects exceed your context window and require periodic reads, error logging, and session catchup. This approach suits multi-step automation tasks, feature development, and workflows needing persistent external memory.

Agents lose track of goals during long automation tasks because context windows fill up. Use a persistent task_plan.md file to refresh goals into the model's attention window before major decisions, and update it after each phase completes to maintain progress.