Plans

You hit your daily credit cap mid-refactor on Tuesday. Wednesday morning, your agent has zero memory of where you stopped. The work-in-progress lives in three things: your last commit, a git stash, and the chat history you can’t fully restore.

Plans replace that. A plan is a tracked piece of work — text, priority, dependencies, optional sub-steps — that lives in .brainclaw/ and travels with the repo. Wednesday morning, your agent reads bclaw_context kind=delta since=<last-session> and sees: this plan is in_progress, these steps are done, this one is the next. No reconstruction.

# Tuesday — agent files what it's working on
bclaw_create entity=plan data={
  text: "Auth rollout — JWT verify",
  priority: "high",
  related_paths: ["src/auth/"]
}

# Wednesday — same or another agent picks it up
bclaw_find entity=plan filter={status: "in_progress"}

Plans support sub-steps, dependency graphs, and lifecycle transitions (todo → in_progress → done | dropped | blocked). Combine with claims before editing, handoffs when you pause, and you have continuous work across as many sessions as it takes.