Planning is a **PM activity**. When a PM (Cell PM or Main PM) picks up a coordination or parent task, they record a plan with `i_will_plan` and then fan the work out into subtasks with `delegate`.
Developers do not have a separate planning verb — they pass a short `plan` argument directly to `i_will_work_on(task_id, plan="...")` when they claim a coding task.
When you decompose a parent task, declare which parent criteria each subtask is responsible for with **`covers_parent_criteria`** (a list of the parent's `acceptance_criteria_ids`). This is what lets the org prove a decomposition covers the parent's full intent — and it drives two gates and your coverage briefing.
- **`parent_ac_coverage`** — one entry per parent criterion: its `id`, `text`, whether a live subtask `claimed` it, and whether a completed subtask `verified` it.
- **`unclaimed_parent_acs`** — the parent criterion ids that no live subtask covers yet. Keep delegating until this is empty.
- **Decomposition floor** — you cannot go `i_am_idle` on a parent while a criterion is still unclaimed. Delegate (or `reassign`) subtasks until every criterion is covered.
- **Roll-up gate** — a parent cannot `complete`, `submit_up`, or `escalate_to_ceo` unless every criterion traces to a child that **passed QA** on it.
Both gates are **safe-by-construction**: they stay inert until you start declaring `covers_parent_criteria`, so a decomposition that never declares coverage is never blocked. Declaring coverage is how you opt your parent into the guarantee.
For code subtasks, delegate each developer their **full queue up front** rather than one task at a time. Both of a cell's developers build in parallel, and each works its own queue one task at a time, in order:
- A per-lane dispatch barrier holds a developer's later subtasks until their current one is in flight — so each dev's lane stays sequenced while the two devs run concurrently.
- Leaf PRs are still merged into the shared cell branch **in sequence**, not in parallel.
A PM (Cell PM or Main PM) is a coordinator — it plans and delegates, it does not write code, and it has no code verb. The role×task_type rule is enforced at **creation**, not just at delegate:
- A `code`-typed task **cannot be assigned to a PM** — `delegate`, `TaskService.create`, batch activation, reassign, and the claim/escalation diversion all consult the same `pm_cannot_own_code` / `main_pm_cannot_own_code` guard and reject it.
- A Main-PM coordination **root** that is `code`-typed is rejected by `submit_root`'s `PRECONDITION_ROOT_NOT_CODE` — a Main PM can never assemble+merge a code root, because it can't have written one.
- The one exception — "a PM may take a code task **only to resolve review issues**" (`is_issue_resolution`) — is a server-side signal the platform sets when routing a `needs_revision` code task back to its owning PM to act on concrete review issues; it is **not** something you pass from a verb. In practice no live path exercises it yet; the structural rule is: if you're a PM and you're looking at a `code` task, delegate it to a developer instead.
This is structural, not a hint. Before this guard, a PM assigned a code task would claim it and deadlock into a respawn loop — a coordinator with no code verb holding a code task it can neither do nor hand back. The guard makes that loop unrepresentable.
## Delegation Depth
The task hierarchy is capped at `MAX_TASK_DEPTH = 4` levels (depths 0–3). The normal 3-layer flow (Main-PM root → cell task → dev subtask) fits in 3; **MegaTask** adds one Main-PM layer on top — umbrella (depth 0) → root-subtask (1) → cell task (2) → dev subtask (3) — which is why the cap is 4, not 3. A `delegate` that would create a node at depth 4 is rejected with a clean `invalid_state` and a "create as a sibling" remediation. Don't over-nest; if you're hitting the cap, the work belongs as a sibling, not a child.