JSON for the current iteration (what to work on now). Preferred agent entry — avoids reading all of PLAN.md.

JSON for one iteration by version (goal, tasks, progress).

List iterations filtered by open | complete | all.

Locate tasks by substring match (JSON).

Backlog section items (JSON).

ASCII view of the resolved current iteration (humans/logs).

Compact index: counts + current iteration. Agents should prefer get_current_iteration / get_iteration over this or full PLAN.md.

Structure lint: duplicates, multiple current, complete-with-open-tasks.

Bootstrap a new PLAN.md when none exists (powernote-style skeleton). Default path: ./PLAN.md in cwd. Refuses overwrite unless force=true.

Create ## vX.Y — Title major section.

Create ### vX.Y.Z — Title iteration.

Set/replace Goal: on an iteration.

Append a checkbox task to an iteration.

Append several checkbox tasks in one write. Prefer this over repeated add_task. Shared done/agent apply to every item.

Tick one or many tasks in one write. For several, pass indexes (preferred after get_iteration) or tasks. For exactly one, index or task still work.

Untick one or many tasks in one write. For several, pass indexes or tasks. For exactly one, index or task still work.

Rewrite task text, preserving done state. One task: text + index/task. Several: changes[{text, index|task, expect?}].

Delete one or many tasks in one write. For several, pass indexes or tasks. Resolve-all then drop (indexes do not shift).

Move one or many tasks to the backlog in one write. For several, pass indexes or tasks. Shared reason/agent apply to every item.

Append one (text) or many (texts) items to Future (Backlog) in one write.

Append freeform markdown at top level.

Mark iteration ACTIVE / current; clears other current markers.

Mark iteration COMPLETE. Requires force=true if open tasks remain. Optional stamp (e.g. date) appended to title.