Back to skills
SKILL.md
Orchestration
ASecurity多任务并行编排调度。已批准 plan +「开始实现」后并行派发 WU。平台无关;物理绑定见 platform/<platform>/bindings.md。触发:并行实现、多 task、开始实现、编排。
- 3 stars
- 0 votes
- 0 copies
- 1 view
- Added September 22, 2026
Works with
Security analysis
100/100npx -y skills add wrysmil/Code-Harness-Kit --skill orchestration --agent claude-codeAre you the author of Orchestration?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/wrysmil-orchestration)---
name: orchestration
description: 多任务并行编排调度。已批准 plan +「开始实现」后并行派发 WU。平台无关;物理绑定见 platform/<platform>/bindings.md。触发:并行实现、多 task、开始实现、编排。
---
# Orchestration(统一编排)
**前置:** 已批准 plan;用户说「开始实现」。未批准不得激活。
**平台:** 本 skill 平台无关。SpawnWorker 具体机制见 `platform/<platform>/bindings.md`。禁用平台原生 plan 工具(见 `routing.md` § 平台原生 plan 工具)。
## 激活后
1. 声明 `「Harness:orchestration:dispatcher-workflow」`
2. Read **`harness-kit/core/orchestration/dispatcher-workflow.md`**
3. Read `tracking/schema.md`、已批准 plan、`project.verification.md`
4. 委派写代码 WU:**WORKTREE-INIT** → 并行派发
## SpawnWorker
各平台子 Agent 委派机制见对应适配器:
| 平台 | 绑定文件 | 机制摘要 |
| --- | --- | --- |
| Claude Code | `platform/claude/bindings.md` | `Task(subagent_type=generalPurpose)` + core agent 文件为 prompt |
| Cursor | `platform/cursor/bindings.md` | `.agents/agents/<role>.md` subagent |
| Trae | `platform/trae/bindings.md` | Trae Agent 模式 + core agent 文件 |
通用 agent_role 映射见 `core/orchestration/dispatcher-workflow.md` § 步骤 2。`wu_skills: auto` → `core/orchestration/skill-preferences.md`。
**尾盘:** collective-test → 并行扇出 reviewer + security-auditor(+ perf-auditor 按需)→ Leader 落盘产物。
## 禁止
- 未过 plan 门禁
- Leader 写业务代码(小改动除外)
- 实现与审查同实例
- 跳过 DISPATCH-TRACK / 尾盘产物
- 末 WU 返回即声称完成
- 使用平台原生 plan 工具(EnterPlanMode / Cursor Plan 模式等)
Attribution
Comments
Loading comments…