director-orchestrate

name	director-orchestrate
description	Runs the Codex Director worker orchestration lane for multi-item or ambiguous bounded assignments. Use when a Director worker brief explicitly invokes $codex-director:director-orchestrate, or when a worker must coordinate subwork, helper/subagent lanes, sequencing, packet handoffs, or evidence reconciliation inside its assigned scope.
metadata	{"short-description":"Coordinate bounded worker lane"}

Director Orchestrate

Quick Start

Use only when explicitly invoked by a Director brief or by the user. This skill is the worker-side orchestration lane, not the parent Director coordinator.

This split workflow skill is worker-side. If its activation marker, skill body, callback text, or readback evidence appears in a parent Director thread, the parent treats it only as routing input or child evidence. It does not load this workflow into the parent and does not permit parent inline browser, repo, provider, or project execution.

1. Read local instructions and the owning Director brief.
2. Load Orchestrate Workflow.
3. Report activation: workflow-skill-loaded:$codex-director:director-orchestrate, selected workflow/playbook, top-level control loop, scope, user-outcome fit, helper/subagent lane plan, direct-leaf status, and evidence contract.
4. Build or update a compact ledger for the bounded assignment.
5. Decompose into the fewest safe items, run helper/subagent lanes for non-trivial work, reconcile evidence, and return concise final evidence.

Native Helper Mapping

During activation, report native helper runtime surface, V2 helper policy, planned RP-style helper profiles, and helper cleanup. multi_agent_v2 is the worker-internal equivalent of RP agent_run/agent_manage, not a substitute for Director-created top-level worker threads.

When V2 is available:

• Treat spawn_agent as RP agent_run op=start: choose a stable lowercase task_name, set agent_type, model, reasoning_effort, service_tier, and fork_turns when exposed, and otherwise state those choices in the helper prompt.
• Treat wait_agent as RP wait/poll: it is only a mailbox wake signal. Use send_message for queued context, followup_task to steer or continue a helper turn, list_agents for status/path checks, and close_agent for cleanup after evidence is consumed.
• Helper briefs must include role/profile, goal, exact scope, leave-alone files/modules, sibling lanes, plan/artifact paths, model/thinking/fork rationale, expected output, evidence standard, and cleanup expectation.
• Fan out only disjoint helper lanes. Name siblings in every parallel brief, wait or poll regularly, process the first completed helper without duplicating in-flight work, and close finished helpers.
• Map RP roles as explore for narrow scouts, pair for complex/default subwork, engineer for clear bounded execution after the plan is known, and design for critique or user-facing polish.
• V2 evidence counts only after this worker reads the helper output, spot-checks material claims, summarizes verified evidence into the ledger, and records close_agent or close_blocked:<reason>.

Workflow

1. Scope and ledger: restate the assignment, done criteria, owning repo/path, constraints, commit authority, worker handle, selected workflow, and the final user outcome this assignment helps prove.
2. Contextualize: identify project nouns, likely files/modules, risks, instructions, and missing facts. Use only narrow direct inspection before delegating deeper context.
3. Decide task shape: direct single-playbook, worker-internal orchestration, or dynamic workflow packetization. Default to orchestration for non-trivial work. If the task mixes browser/desktop flow, provider config, logs, code, deploys, auth, secrets, or final E2E proof, stop and ask the Director to use dynamic workflow or separate goal-bearing workers instead of absorbing the whole incident.
4. Research: run a scout/helper lane for unknowns that affect the plan. Summarize confirmed facts, uncertainty, and plan constraints.
5. Decompose: create up to five work items, preferably two or three. Each item needs done criteria, owner, workflow/playbook, helper lane, dependencies, review gate, and evidence.
6. Plan review: challenge item boundaries, dependency order, parallel safety, risk, tests, and acceptance criteria before execution continues.
7. Execute or staff lanes: perform only the work authorized by the brief. If coordinator-only, return packet briefs/checkpoints instead of implementing.
8. Reconcile: verify each item, read helper outputs, resolve conflicts, update ledger/artifacts, and decide accepted, blocked, stale, or insufficient evidence.
9. Complete: audit readback/evidence, helper/direct-leaf acceptance, review/oracle state, cleanup/archive state, and next action before final.

Escalate Or Stop

Ask the Director to reroute when the work needs separate top-level Codex workers, approval gates, production/external writes, cross-repo integration, long-lived monitors, or .workflow/<slug>/ artifacts. Stop instead of continuing when scope, authority, secrets, or destructive risk is unclear.

Required Invariants

• Stay inside the repo/path, done criteria, commit authority, and constraints in the Director brief.
• Name director-orchestrate and the top-level control loop in activation, checkpoints, and final evidence.
• Non-trivial work needs a real helper/subagent lane or a clear blocked helper capability. Direct leaf is allowed only with separate tiny, mechanical, and low-risk rationale.
• A coordinator checkpoint is not task completion. It must include next-packet-dispatched, monitor-scheduled, blocked-on-dispatch:<reason>, or awaiting-approval:<reason>.
• Do not report the user's goal complete from a prerequisite checkpoint. Name the final verification surface, checkpoint evidence proved here, and remaining proof before Director completion.
• Do not become a monolithic incident executor. Browser/desktop operation, provider/dashboard repair, app-code patching, direct deploy/alias work, security review, and final E2E proof need separate packets/workers when more than one is in scope.
• Do not create nested top-level Codex workers unless the Director explicitly delegates that authority.
• Treat callbacks and final-looking messages as wake signals for Director readback, not acceptance.
• Do not treat wait_agent, list_agents, or helper final-status notifications as accepted item evidence.
• Keep visible updates to final evidence, blockers, decisions, or ownership-changing handoffs.

Output

Return the ledger-relevant facts: work items, worker/helper handles if any, files/artifacts touched, verification, review/oracle status, native helper surface, V2 helper paths/evidence/owner verification/cleanup when used, unresolved risks, cleanup/archive state, and next action.