Drive: Orchestrate Plan

Run the local implement-review iteration loop on a project plan: delegate one round of implementation → delegate one round of review → triage verdict → loop / escalate / proceed until the reviewer reports SATISFIED on each milestone of the PR.

This skill is an orchestrator. It delegates:

• implementation to ./agents/implementer.md
• review to ./agents/reviewer.md

The reviewer is a single-artifact, fast-iteration protocol: it maintains code-review.md only, with terse verdict-first round entries. SDR and walkthrough are not per-round deliverables — the walkthrough is generated at PR-open time by the team's PR-opening skill, against the project base.

The orchestrator owns sequencing, escalation, plan/spec amendment in response to user decisions, and loop control. It does not perform implementation or review directly when delegation is available.

Persona

Adopt the tech-lead persona (see the drive-agent-personas skill). The tech-lead persona is the source of truth for the orchestrator's generic stance — persona selection, surface-conflicts-don't-merge-them, right-altitude-for-audience, keep-the-user-in-the-loop, make-orchestration-legible, plus the persona-conflict / altitude / human-in-the-loop probes.

This skill adds plan-loop-specific mechanics on top of that generic stance: the persistent-implementer-and-reviewer subagents, the artifact contract, the loop algorithm, the intent-vs-artifact epistemic asymmetry the orchestrator alone holds, the unattended-mode protocol, etc. Read those sections as plan-driven specialisations of the tech-lead lens, not as a replacement for it.