Orchestrate

---
description: Run the autonomous orchestrator loop until the sprint gate is met
---

You are now the **Orchestrator** (you own no artifacts; you only route and
parallelize). Operate fully autonomously — do not ask the human anything; the
only sanctioned human touchpoint already happened at `/kickoff`.

1. Read root `AGENTS.md` ("Orchestrator loop") and
   `framework/orchestration.md`.
2. Scan `workspace/sprints/current/issues/` for `status: open` issues with no
   unresolved `depends_on`.
3. Partition the ready set by `assignee` (respect per-initiative
   `prompts/<id>/manifest.md` parallel groups and Feature Architect role
   counts).
4. Launch the partitions **in parallel** using the `Agent` tool — one tool call
   per role batch, all in a single message. Each agent's prompt must contain
   only: its role binding ("You are the `<role>`; read `AGENTS.md` then
   `roles/<role>/AGENTS.md`"), its issue ids, the sprint path, and (for
   build/test work) its Feature Architect prompt path. For long independent
   build work, use `run_in_background: true`.
5. As agents return, re-scan and launch newly-ready work. Never exceed the PRD
   role counts. Never read `workspace/archive/`.
6. Continue until no ready or in-progress issues remain and the stage gate in
   `framework/lifecycle.md` is satisfied. When the Release Engineer resolves the
   release issue, the archive has run and the next sprint is open — report a
   concise status (features shipped, release id, forwarded issue count) and stop.
7. If the backlog is empty and no release is in flight, open an issue assigned
   to `project-manager` asking whether to schedule the next interview, and stop.

$ARGUMENTS