Back to Details

addy-orchestrator

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Addy Orchestrator

Addy编排器

Skills path: 
~/.claude/plugins/cache/addy-agent-skills/agent-skills/1.0.0/skills/
技能路径: 
~/.claude/plugins/cache/addy-agent-skills/agent-skills/1.0.0/skills/

Core Principle

核心原则

This skill never loses focus. At every step, show the user exactly where they are, what just finished, and what's next. Always read the state file first. Always write the state file after. Never move to the next phase without explicit user confirmation.
本技能全程不会偏离目标。每一步都会明确告知用户当前所处阶段、已完成的内容、下一步的计划。始终先读取状态文件，操作完成后始终写入状态文件。未获得用户明确确认前，不会进入下一个阶段。

Phase 0: Intake

阶段0：信息收集

Before doing anything, check for an existing session state file:
<project-dir>/.addy-session.md
If it exists, read it and resume from the last completed phase. Tell the user:
"Resuming: you're on phase N/M — [phase name]. Ready to continue?"
If no state file, interview the user with exactly these questions:
  1. What are you building or fixing? (one sentence)
  2. Does a spec already exist, or are we starting from scratch?
  3. Greenfield or existing codebase?
  4. What does "done" look like?
Then determine the phase sequence (see table below) and confirm it with the user before proceeding.
执行任何操作前，先检查是否存在已有会话状态文件：
<project-dir>/.addy-session.md
如果文件存在，读取文件内容并从上次完成的阶段恢复执行，告知用户：
"正在恢复会话：你当前处于第N/M阶段 — [阶段名称]。是否准备好继续？"
如果没有状态文件，严格按照以下问题向用户收集信息：
  1. 你要构建或修复的内容是什么？（用一句话描述）
  2. 需求规格说明是否已经存在，还是我们需要从零开始梳理？
  3. 是全新项目开发，还是基于现有代码库迭代？
  4. 「完成」的判定标准是什么？
之后确定阶段执行序列（参见下表），在继续前先和用户确认序列是否正确。

Phase Sequences

阶段执行序列

ScenarioSequence
Vague ideaidea-refine → spec-driven-development → planning-and-task-breakdown → incremental-implementation → test-driven-development → code-review-and-quality → git-workflow-and-versioning
Clear spec, no planplanning-and-task-breakdown → incremental-implementation → test-driven-development → code-review-and-quality → git-workflow-and-versioning
Bug fixdebugging-and-error-recovery → test-driven-development → code-review-and-quality
UI featurespec-driven-development → frontend-ui-engineering → browser-testing-with-devtools → code-review-and-quality
API designspec-driven-development → api-and-interface-design → test-driven-development → code-review-and-quality
Security / perfsecurity-and-hardening / performance-optimization (standalone)
Shippingshipping-and-launch
场景执行序列
模糊的想法idea-refine → spec-driven-development → planning-and-task-breakdown → incremental-implementation → test-driven-development → code-review-and-quality → git-workflow-and-versioning
有明确的需求规格说明，但没有执行计划planning-and-task-breakdown → incremental-implementation → test-driven-development → code-review-and-quality → git-workflow-and-versioning
缺陷修复debugging-and-error-recovery → test-driven-development → code-review-and-quality
UI功能开发spec-driven-development → frontend-ui-engineering → browser-testing-with-devtools → code-review-and-quality
API设计spec-driven-development → api-and-interface-design → test-driven-development → code-review-and-quality
安全/性能优化security-and-hardening / performance-optimization (独立执行)
版本发布shipping-and-launch

State File Format

状态文件格式

Write 
.addy-session.md
in the project directory after every phase:
markdown
undefined
每个阶段执行完成后，在项目目录下写入 
.addy-session.md
文件：
markdown
undefined

Addy Session

Addy Session

task: <one-line description> started: <date> sequence: [phase1, phase2, phase3, ...] current_phase: <index starting at 0> status: in_progress | complete
task: <一行描述任务> started: <日期> sequence: [阶段1, 阶段2, 阶段3, ...] current_phase: <从0开始的索引> status: in_progress | complete

Completed Phases

已完成阶段

phase-name

阶段名称

  • Summary: ...
  • Artifacts: ...
  • Decisions: ...
  • 摘要: ...
  • 产出物: ...
  • 决策: ...

Accumulated Context

累计上下文

<running context block — grows with each phase>

---
<运行时上下文块 — 随每个阶段执行不断补充>

---

Phase Execution Loop

阶段执行循环

Repeat this loop for every phase until all are done:
对每个阶段重复执行以下循环，直到所有阶段全部完成：

1. Announce the phase

1. 宣告当前阶段

Always show a status header:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Phase N/M: [PHASE-NAME]
Done: ✓ phase1  ✓ phase2
Next: phase4, phase5
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
始终先展示状态头部：
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Phase N/M: [阶段名称]
已完成: ✓ phase1  ✓ phase2
下一步: phase4, phase5
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

2. Spawn the subagent

2. 生成子Agent

You are executing the [SKILL_NAME] phase of a structured development workflow.

Read and follow the skill at:
~/.claude/plugins/cache/addy-agent-skills/agent-skills/1.0.0/skills/[SKILL_NAME]/SKILL.md

== Accumulated context ==
[paste context block from state file]

== Your task ==
[specific instruction for this phase]

When done, output:
PHASE SUMMARY: <2-3 sentences>
ARTIFACTS: <list of files created/modified>
DECISIONS: <any choices made that affect later phases>
BLOCKERS: <anything requiring human input before next phase>
You are executing the [SKILL_NAME] phase of a structured development workflow.

Read and follow the skill at:
~/.claude/plugins/cache/addy-agent-skills/agent-skills/1.0.0/skills/[SKILL_NAME]/SKILL.md

== Accumulated context ==
[paste context block from state file]

== Your task ==
[specific instruction for this phase]

When done, output:
PHASE SUMMARY: <2-3 sentences>
ARTIFACTS: <list of files created/modified>
DECISIONS: <any choices made that affect later phases>
BLOCKERS: <anything requiring human input before next phase>

3. Process the result

3. 处理执行结果

  • Extract PHASE SUMMARY / ARTIFACTS / DECISIONS / BLOCKERS from subagent output
  • Update the state file
  • If BLOCKERS is non-empty: stop and resolve with the user before continuing
  • Show the user the phase result, then ask: "Ready to move to [next phase], or do you want to adjust anything first?"
  • Do not auto-advance — always wait for explicit go-ahead
  • 从子Agent的输出中提取阶段摘要/产出物/决策/阻塞项
  • 更新状态文件
  • 如果阻塞项非空：停止执行，先和用户一起解决阻塞问题后再继续
  • 向用户展示阶段执行结果，然后询问："是否准备好进入[下一阶段]，还是你想要先调整相关内容？"
  • 不要自动进入下一阶段 — 始终等待用户明确同意后再推进

4. Never abandon the loop

4. 永远不要跳出循环

If the user goes on a tangent, acknowledge it, then redirect:
"Got it. Let me note that for phase N. To keep the workflow on track — want to continue with [current phase] or address this first?"
如果用户话题跑偏，先确认用户的需求，然后引导回归流程：
"了解了，我会把这个需求记录到第N阶段的待办中。为了保证工作流按计划推进，你想要继续执行[当前阶段]，还是先处理这个需求？"

Completion

执行完成

When all phases are done:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✓ All phases complete
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Present:
  • What was built/fixed (1 paragraph)
  • Files changed or created
  • Decisions made along the way
  • Suggested next steps (deploy, monitor, doc)
Mark 
.addy-session.md
status as 
complete
.
所有阶段全部完成后：
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✓ 所有阶段执行完成
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
向用户展示以下内容：
  • 本次构建/修复的内容（1段描述）
  • 变更或新增的文件列表
  • 执行过程中做出的所有决策
  • 建议的下一步操作（部署、监控、文档编写）
将 
.addy-session.md
文件的状态标记为 
complete
。