workflow

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Workflow Setup

工作流安装

Installs the development workflow layer. Requires an existing

.claude/

directory — run the greenfield or brownfield skill first.

安装开发工作流层。需要已存在的

.claude/

目录 ——请先运行greenfield或brownfield技能。

Step 1 — Preflight

步骤1 — 预检

Check that prerequisites exist:

bash

ls .claude/ 2>/dev/null || echo "MISSING"
ls .claude/hooks/stop-quality-gate.sh 2>/dev/null && echo "GATE_EXISTS" || echo "GATE_MISSING"
cat .claude/settings.json 2>/dev/null | grep -q "AGENT_TEAMS" && echo "TEAMS_ENABLED" || echo "TEAMS_DISABLED"

.claude/

is missing, stop:

"No .claude/ directory found. Run the greenfield or brownfield skill first, then re-run /workflow."

.claude/commands/

.claude/agents/

already have files, warn:

"Existing commands/agents will be overwritten. Continue? (yes/no)"

检查是否存在必要条件：

bash

ls .claude/ 2>/dev/null || echo "MISSING"
ls .claude/hooks/stop-quality-gate.sh 2>/dev/null && echo "GATE_EXISTS" || echo "GATE_MISSING"
cat .claude/settings.json 2>/dev/null | grep -q "AGENT_TEAMS" && echo "TEAMS_ENABLED" || echo "TEAMS_DISABLED"

如果缺少

.claude/

目录，请停止：

"未找到.claude/目录。请先运行greenfield或brownfield技能，然后重新运行/workflow。"

如果

.claude/commands/

或

.claude/agents/

已存在文件，请发出警告：

"现有commands/agents将被覆盖。是否继续？(是/否)"

Step 2 — Check for Previous Configuration

步骤2 — 检查先前配置

Before starting the interview, check if this skill has been run before:

bash

SKILL_NAME="workflow" \
bash "${CLAUDE_SKILL_DIR}/scripts/load-config.sh"

If the output is a non-empty JSON object (not

{}

), previous answers exist. Present them:

Previous workflow configuration found:

  Orchestration: [ORCHESTRATION]
  Commands:      [COMMANDS]
  Gitleaks:      [GITLEAKS]
  Semgrep:       [SEMGREP]

Would you like to:
  a) Update — reinstall all commands and agents using these settings
  b) Reconfigure — start the interview from scratch

Use

AskUserQuestion

and wait for the response.

If Update: load all saved values into environment variables and skip to Step 3.
If Reconfigure: proceed to the full interview below.

If no previous config exists, proceed directly to the interview.

开始交互前，检查此技能是否已运行过：

bash

SKILL_NAME="workflow" \
bash "${CLAUDE_SKILL_DIR}/scripts/load-config.sh"

如果输出为非空JSON对象（不是

{}

），说明存在先前的配置答案。展示这些内容：

找到先前的工作流配置：

  编排模式: [ORCHESTRATION]
  命令:      [COMMANDS]
  Gitleaks:      [GITLEAKS]
  Semgrep:       [SEMGREP]

您希望：
  a) 更新 — 使用这些设置重新安装所有命令和Agent
  b) 重新配置 — 从头开始进行交互配置

使用

AskUserQuestion

并等待响应。

如果选择更新：将所有保存的值加载到环境变量中，并跳至步骤3。
如果选择重新配置：继续进行以下完整的交互配置。

如果没有先前的配置，则直接进入交互配置环节。

Step 2b — Interview

步骤2b — 交互配置

Ask each question using

AskUserQuestion

and wait for the response before proceeding to the next question.

使用

AskUserQuestion

逐个提问，等待每个问题的响应后再进行下一个问题。

Question 1 — Orchestration Mode

问题1 — 编排模式

Use

AskUserQuestion

with this prompt:

Setting up your development workflow.

Orchestration mode — How should /breakdown, /spec, and /work orchestrate agents?

  a) Subagents — parallel (recommended) — parallel workers in isolated git
     worktrees, merged by the git-expert agent. Stable, works everywhere.

  b) Subagents — PEE loop — sequential Planner → Executor → Evaluator loop
     per spec on a shared branch, with one full quality pass at the end.
     One-shot subagents (no team infrastructure). Optimized for spec-compliance
     correctness, not raw speed. Uses opus for planner/evaluator and sonnet
     for the executor.

  c) Agent teams (experimental) — teammates work on parallel domains of the
     codebase on a shared branch. Requires CC v2.1.32+ and
     CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS enabled.

Map the answer to the internal

ORCHESTRATION

value: a →

subagents

, b →

loop

, c →

teams

使用

AskUserQuestion

并传入以下提示：

正在设置您的开发工作流。

编排模式 — /breakdown、/spec和/work应如何编排Agent？

  a) 子Agent — 并行（推荐）—— 在独立的git工作树中并行处理，由git-expert Agent合并。稳定，适用于所有场景。

  b) 子Agent — PEE循环 — 在共享分支上按规格依次执行规划器→执行器→评估器循环，
     最后进行一次完整的质量检查。单次调用子Agent（无需团队基础设施）。针对规格合规性优化，
     而非追求速度。规划器/评估器使用opus模型，执行器使用sonnet模型。

  c) Agent团队（实验性）—— 团队成员在共享分支上并行处理代码库的不同领域。需要CC v2.1.32+版本，
     且需启用CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS。

将答案映射为内部

ORCHESTRATION

值：a →

subagents

，b →

loop

，c →

teams

。

Question 2 — Commands

问题2 — 命令选择

Use

AskUserQuestion

with this prompt:

Which commands would you like installed? (default: all)

  a) /breakdown, /spec, /work  — feature planning and implementation   ✓
  b) /commit                   — conventional commit message generator  ✓
  c) /review                   — code review of current changes         ✓
  d) /pr, /squash-pr, /address-pr-comments — pull request workflow      ✓
  e) /fix-issue                — implement a GitHub issue end-to-end    ✓
  f) /security-scan            — on-demand deep security scan           ✓
  g) /triage                   — GitHub issue analysis (requires gh)    ✓
  h) /update-skills            — find and install skills for your deps  ✓

Enter "all" or list letters to include (e.g. "a, b, d"):

使用

AskUserQuestion

并传入以下提示：

您希望安装哪些命令？（默认：全部）

  a) /breakdown、/spec、/work  — 功能规划与实现   ✓
  b) /commit                   — 规范化提交消息生成器  ✓
  c) /review                   — 当前变更的代码评审         ✓
  d) /pr、/squash-pr、/address-pr-comments — 拉取请求工作流      ✓
  e) /fix-issue                — 端到端实现GitHub问题    ✓
  f) /security-scan            — 按需深度安全扫描           ✓
  g) /triage                   — GitHub问题分析（需要gh）    ✓
  h) /update-skills            — 为项目依赖查找并安装技能  ✓

输入"all"或列出要包含的选项字母（例如"a, b, d"）：

Step 2c — Save Configuration

步骤2c — 保存配置

After the interview (or after loading previous config for an update), save the current answers:

bash

SKILL_NAME="workflow" \
CONFIG_JSON='{"ORCHESTRATION":"[val]","COMMANDS":"[val]","GITLEAKS":"[val]","SEMGREP":"[val]"}' \
bash "${CLAUDE_SKILL_DIR}/scripts/save-config.sh"

Replace

[val]

with actual values. This writes to

.claude/skill-config.json

so future runs can reuse these answers.

完成交互配置（或加载先前配置进行更新）后，保存当前答案：

bash

SKILL_NAME="workflow" \
CONFIG_JSON='{"ORCHESTRATION":"[val]","COMMANDS":"[val]","GITLEAKS":"[val]","SEMGREP":"[val]"}' \
bash "${CLAUDE_SKILL_DIR}/scripts/save-config.sh"

将

[val]

替换为实际值。此操作会将配置写入

.claude/skill-config.json

，以便后续运行时可以复用这些答案。

Step 3 — Detect Quality Tools

步骤3 — 检测质量工具

bash

bash scripts/detect-tools.sh

For any missing tools, list what they do and how to install them. Do NOT offer to install them automatically.

Recommended tools:

  gitleaks — Scans for accidentally committed secrets
    https://github.com/gitleaks/gitleaks#installing
    brew install gitleaks

  semgrep — SAST security scanner (injection, OWASP Top 10)
    https://semgrep.dev/docs/getting-started/
    pip install semgrep  /  brew install semgrep

  gh CLI — Required for /triage, /pr, /squash-pr, /address-pr-comments
    https://cli.github.com/
    brew install gh

The quality gate will skip checks for tools that aren't installed.

bash

bash scripts/detect-tools.sh

对于任何缺失的工具，列出其功能和安装方法。请勿自动安装这些工具。

推荐工具：

  gitleaks — 扫描意外提交的敏感信息
    https://github.com/gitleaks/gitleaks#installing
    brew install gitleaks

  semgrep — SAST安全扫描器（检测注入攻击、OWASP Top 10等）
    https://semgrep.dev/docs/getting-started/
    pip install semgrep  /  brew install semgrep

  gh CLI — /triage、/pr、/squash-pr、/address-pr-comments命令所需
    https://cli.github.com/
    brew install gh

对于未安装的工具，质量门会跳过相应检查。

Step 4 — Install Commands and Agents

步骤4 — 安装命令和Agent

Run in parallel if subagents are available, otherwise sequential:

bash

ORCHESTRATION="[subagents|teams|loop]" \
COMMANDS="[comma-separated approved list]" \
bash scripts/install-commands.sh

ORCHESTRATION="[subagents|teams|loop]" \
bash scripts/install-agents.sh

If gh CLI is available and /triage was approved:

bash

bash scripts/install-triage-skill.sh

如果支持子Agent则并行运行，否则顺序运行：

bash

ORCHESTRATION="[subagents|teams|loop]" \
COMMANDS="[逗号分隔的已批准列表]" \
bash scripts/install-commands.sh

ORCHESTRATION="[subagents|teams|loop]" \
bash scripts/install-agents.sh

如果gh CLI已安装且/triage命令已获批：

bash

bash scripts/install-triage-skill.sh

Step 5 — Verify Quality Gate

步骤5 — 验证质量门

Confirm the installed

.claude/hooks/stop-quality-gate.sh

is the scoped, stack-aware template (gitleaks and semgrep are now built into the hook's shared security block and run only on changed files — no patching needed):

bash

bash scripts/patch-quality-gate.sh

The script is a no-op when the hook is already the scoped template. If it detects an older unscoped hook, it prints instructions to re-run greenfield/brownfield to regenerate.

确认已安装的

.claude/hooks/stop-quality-gate.sh

是支持作用域、感知栈的模板（gitleaks和semgrep现已内置到钩子的共享安全模块中，仅针对变更文件运行——无需补丁）：

bash

bash scripts/patch-quality-gate.sh

如果钩子已经是作用域模板，该脚本不会执行任何操作。如果检测到旧版非作用域钩子，它会打印说明，提示重新运行greenfield/brownfield技能以重新生成。

Step 6 — Patch Settings for Agent Teams (if chosen)

步骤6 — 为Agent团队修补设置（若选择该模式）

If orchestration =

teams

only (the PEE loop mode uses one-shot subagents and does not need team infrastructure):

bash

bash scripts/patch-settings-teams.sh

Adds

TaskCompleted

hook to

.claude/settings.json

that runs

format.sh

and

task-summary.sh

. Quality gates are handled by the dedicated Quality Gate agent between groups — no

teammate-quality-gate.sh

is created.

Note to developer after patching:

VS Code may show schema warnings on
TaskCompleted
. This is a known schema lag — agent teams hooks are newer than the schema. The hooks work correctly at runtime. To suppress the warning add this to
.vscode/settings.json
:
{ "json.schemas": [{ "fileMatch": ["**/.claude/settings.json"], "schema": {} }] }

仅当编排模式为

teams

时执行此步骤（PEE循环模式使用单次调用子Agent，不需要团队基础设施）：

bash

bash scripts/patch-settings-teams.sh

在

.claude/settings.json

中添加

TaskCompleted

钩子，该钩子会运行

format.sh

和

task-summary.sh

。质量检查由专门的Quality Gate Agent在各组之间处理——不会创建

teammate-quality-gate.sh

。

修补后给开发者的提示：

VS Code可能会对
TaskCompleted
显示架构警告。这是已知的架构滞后问题——Agent团队钩子比架构更新。钩子在运行时可以正常工作。要抑制警告，请将以下内容添加到
.vscode/settings.json
中：
{ "json.schemas": [{ "fileMatch": ["**/.claude/settings.json"], "schema": {} }] }

Step 7 — Commit

步骤7 — 提交

bash

git add .claude/
git commit -m "chore(claude): install development workflow"

bash

git add .claude/
git commit -m "chore(claude): install development workflow"

Step 8 — Completion

步骤8 — 完成

✅ Workflow Setup Complete

Commands installed:
  /breakdown + /spec + /work  — [subagents parallel | subagents PEE loop | agent teams] orchestration
  /commit                     — conventional commit message generator
  /review                     — code review of current changes
  /pr + /squash-pr            — pull request workflow
  /address-pr-comments        — address GitHub review comments
  /fix-issue                  — implement a GitHub issue end-to-end
  /security-scan              — on-demand deep security scan
  /triage                     — GitHub issue analysis and planning
  /update-skills              — find and install skills for project dependencies

Agents installed:
  architect                                       — plans features and architecture
  [implementer | domain-implementer | planner+executor+evaluator]  — implement specs
  quality                                         — 8-gate quality review
  git-expert                                      — worktrees, merges, verification

Quality tools:
  gitleaks: [✅ active in quality gate | ⚠️ not installed]
  semgrep:  [✅ active in quality gate | ⚠️ not installed]
  gh CLI:   [✅ installed | ⚠️ not installed — /triage, /pr, /squash-pr need this]

Agent teams hooks: [added to settings.json | not needed for this orchestration mode]

⚠️ Restart Claude Code to load the new commands and agents.

Typical workflow once restarted:
  /breakdown my-feature     — decompose into tasks
  /spec my-feature          — generate implementation specs
  /work my-feature --all    — implement with agents
  /squash-pr my-feature     — create PR

✅ 工作流安装完成

已安装的命令：
  /breakdown + /spec + /work  — [子Agent并行 | 子Agent PEE循环 | Agent团队] 编排模式
  /commit                     — 规范化提交消息生成器
  /review                     — 当前变更的代码评审
  /pr + /squash-pr            — 拉取请求工作流
  /address-pr-comments        — 处理GitHub评审意见
  /fix-issue                  — 端到端实现GitHub问题
  /security-scan              — 按需深度安全扫描
  /triage                     — GitHub问题分析与规划
  /update-skills              — 为项目依赖查找并安装技能

已安装的Agent：
  architect                                       — 规划功能与架构
  [implementer | domain-implementer | planner+executor+evaluator]  — 实现规格要求
  quality                                         — 8关卡质量评审
  git-expert                                      — 工作树管理、合并、验证

质量工具状态：
  gitleaks: [✅ 在质量门中激活 | ⚠️ 未安装]
  semgrep:  [✅ 在质量门中激活 | ⚠️ 未安装]
  gh CLI:   [✅ 已安装 | ⚠️ 未安装 — /triage、/pr、/squash-pr命令需要此工具]

Agent团队钩子：[已添加到settings.json | 此编排模式不需要]

⚠️ 请重启Claude Code以加载新的命令和Agent。

重启后的典型工作流：
  /breakdown my-feature     — 拆解任务
  /spec my-feature          — 生成实现规格
  /work my-feature --all    — 通过Agent实现功能
  /squash-pr my-feature     — 创建PR

Reference Files

参考文件

Commands:
```
references/commands/
```
Agents:
```
references/agents/
```
Triage skill:
```
references/triage-skill.md
```

命令：
```
references/commands/
```
Agent：
```
references/agents/
```
问题分类技能：
```
references/triage-skill.md
```