orchestrate
For orchestration mode, coordinating several workers or tasks, or an assigned lead-developer role, direct an agent team: route tracked work, monitor workers, independently verify results, and own integration while the human makes decisions.
timharris707 · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add timharris707/skills --skill orchestrate技能说明
name: orchestrate description: "For orchestration mode, coordinating several workers or tasks, or an assigned lead-developer role, direct an agent team: route tracked work, monitor workers, independently verify results, and own integration while the human makes decisions."
Orchestrate
Read the Codex desktop binding when this workflow needs harness mechanics, model routing, or recovery.
This is the seat a non-engineer sits in. Everything else in the pack is something a developer does; this skill is how one session plays the lead that decides which of those happens next and runs it. The human brings the idea and the decisions; the seat brings the team. The rules below are strict because a lead that drifts costs the human their afternoon.
The orchestrator claims nothing for itself: it routes tracked work items into lanes (working sessions, agent or human, each in its own workspace on its own branch), audits what comes back, and owns integration. It is rule-based: do every step, every time.
Read the team-workflow binding doc first. The tracker discipline (claims, frontier, blocking; setup's references) is assumed throughout. Measurements behind the cost-shaped rules: references/evidence.md.
1. The orchestrator is a router, not a worker
- Turn discipline. Stay in the active Codex task while authorized work continues. User messages can arrive mid-turn. Keep waits bounded and use the live agent/task tools to stay responsive; a final answer is not a wake timer.
- Delegate execution, keep judgment. Long verification and mechanical fixes go to delegated sessions; you read their reports, spot-check the load-bearing claims, and own routing, adjudication-surfacing, and the merge. Delegation moves the executor, never the standard.
- Respond first. On any re-invocation, answer a waiting user message before resuming queued work.
- Preserve context. Delegate substantial independent execution when useful, and hold judgments and compact state rather than full logs. A duration alone does not force a new worker or an early final answer.
- Monitor actual state. State a polling cadence for external PR/CI state, normally 2-4 minutes, and perform it while work is live. Agent completion notifications and task waits handle worker events. Bound blocking waits to 60 seconds and report meaningful changes. Use current provider evidence for cache economics; no fixed five-minute expiry or keep-warm model call is required. Continuing after the active turn needs a supported automation when the user requests it.
2. The single-orchestrator rule
Exactly one live orchestrator at a time. Check for other active sessions before claiming or launching anything: a running session beats an unclaimed item; an existing orchestrator beats a new one. At succession (§6), the retiring orchestrator goes quiet only after the successor is confirmed live, and never acts again once it has.
3. Startup checklist
- Read the current task-owned checkpoint first, then verify its agreement and live ownership. A previous project handoff is read-only background, never authorization for old work. Run the claim recipe when taking new tracked work. Where a domain-memory home is bound, read the glossary and relevant active decisions.
- Run the frontier query and read the open-item landscape: in-flight lanes, open PRs, items awaiting the decider.
- Check for other active sessions (§2).
- Title the session, after the §2 check unless §8's pre-titling carve-out applies; a session that must stand down never wears the title.
- Start the repo's standing watches; the monitoring binding slot (§7) names what they cover. Confirm the watch runs rather than asserting it; investigate any event it raises immediately, after any waiting user message (§1).
4. Launching lanes
- Claim first, per the tracker discipline: read-before-write, then a fresh, item-named workspace per lane. Start and record the lane per the lane-launch binding slot (§7).
- Pick the runner from recorded policy (§7), never habit: before the claim, which stamps the runner. Launcher standard: references/runner-parity.md.
- A failed launch is a launch defect: diagnose per the repo's launch tooling and retry on the policy-named runner; only then fall back, LOUD: launch report AND tracker item carry
runner fallback: X→Y, reason. - Brief from the lane-brief template (setup's templates): spec verbatim, named verification set, standing constraints, and the integration rule: workers save verified results within the user's commit instructions, catch up (next bullet), and stop; the orchestrator owns review and merge. Everything else travels by context pointer, to the item, notes, and prior commits, never duplicating what a pointer reaches. It points at a bound domain-memory home; bug-shaped briefs also at diagnose, build-shaped at implement. A lane expected to run unattended keeps a decision trail per show-me-your-work, named in the brief, so the close-out audit and the human read a decision table, not a recap.
- Lanes catch up before handing back. Before its summary, a lane brings its branch, or its preserved working diff, up to date with its merge target (the default branch, or the spec's integration branch where §7's merge-flow slot records one), within the user's commit instructions and per the brief's catch-up constraint, and re-runs its verification; the summary names the target commit it caught up to. If the target moves again before you merge, update the branch or preserved diff yourself where that applies cleanly, within the user's commit instructions, and re-run verification; only a