# i moved my project plan out of CLAUDE.md and it stopped forgetting stuff around turn 20 — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/a8c560cc12b340e3

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-11-13 04:59 ingested 2026-11-14 02:41

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xuyyo3/i_moved_my_project_plan_out_of_claudemd_and_it/)

## retained text



```
CLAUDE.md gets loaded every turn, so as my plan grew every message was carrying the whole thing, including tasks that were done weeks ago. long sessions got heavy and claude still lost track of earlier decisions anyway.

what i do now: CLAUDE.md is tiny, basically the project rules plus "fetch your next task". the plan lives outside the repo as tasks with dependencies, and each task only carries the constraints it actually touches. claude pulls the next unblocked task, does it, reports what it changed, and then i /clear. every session starts small instead of dragging the whole project along.

stuff i learned doing this: vague tasks are where it falls apart. cheaper models do fine on a tight task and badly on a loose one, so most of the effort goes into writing the task well. and dependencies matter more than i thought, without them it'll happily start task 4 before task 2 exists.

disclosure, the task server i use is my own thing (an MCP called nodeon), but you can get most of this with one markdown file per task and a line in CLAUDE.md telling it to only work the first unchecked one. curious how other people are handling plans on bigger projects
```



document fetched 2026-11-14 00:45 · canonical https://reddit.com/r/ClaudeCode/comments/1xuyyo3/i_moved_my_project_plan_out_of_claudemd_and_it
