planning-with-files

Manage multi-step AI workflows by creating persistent on-disk planning files like task_plan.md, findings.md, and progress.md. This external planning workspace keeps goals, progress, and discoveries accessible across sessions without relying on internal memory.

The best way to plan complex automation projects is using a file-based planning workspace. Defining an end state, current phase, and planned phases in task_plan.md structures tasks requiring many tool calls and prevents context loss.

Create a structured findings log by maintaining a findings.md file in your project directory. This tracks discoveries and research output externally, providing a persistent record that survives interruptions and supports multi-phase task execution.

Yes, you can use bash, grep, date, and cat to interact with on-disk planning artifacts. These dependencies support managing session progress logs and task_plan.md files for multi-phase projects requiring structured, persistent tracking.

You should not rely on internal memory for task management when tasks require more than five tool calls or span multiple sessions. Complex tasks overwhelm internal memory, making persistent on-disk planning files necessary for robust recovery.

Planning with external files works for recovering AI workflows after interruptions. It keeps planning context external to the AI interaction through progress.md and task_plan.md, enabling easier recovery and robust multi-step workflows.