delegated-execution

委派执行

通过为每项任务派遣一名全新的子 Agent 来执行计划，并在每项任务完成后进行两阶段审查：先检查规格符合性，再检查代码质量。

核心原则： 每项任务都使用全新的子 Agent ＋ 两阶段审查（先检查规格，再检查质量）＝ 高质量、快速迭代

适用场景

digraph applicability {
    "Implementation plan exists?" [shape=diamond];
    "Tasks mostly independent?" [shape=diamond];
    "Stay in current session?" [shape=diamond];
    "delegated-execution" [shape=box];
    "task-runner" [shape=box];
    "Manual execution or go back to intent-discovery" [shape=box];

    "Implementation plan exists?" -> "Tasks mostly independent?" [label="yes"];
    "Implementation plan exists?" -> "Manual execution or go back to intent-discovery" [label="no"];
    "Tasks mostly independent?" -> "Stay in current session?" [label="yes"];
    "Tasks mostly independent?" -> "Manual execution or go back to intent-discovery" [label="no - tightly coupled"];
    "Stay in current session?" -> "delegated-execution" [label="yes"];
    "Stay in current session?" -> "task-runner" [label="no - separate session"];
}

与任务运行器（独立会话）相比：

• 在同一会话中进行（无需切换上下文）
• 每项任务都使用全新的子 Agent（无上下文污染）
• 每项任务完成后进行两阶段审查：先检查规格符合性，再检查代码质量
• 迭代更快（任务之间无需人工介入）

流程

digraph workflow {
    rankdir=TB;

    subgraph cluster_per_task {
        label="Per Task";
        "Dispatch builder subagent (./implementer-prompt.md)" [shape=box];
        "Builder subagent has questions?" [shape=diamond];
        "Answer questions, supply context" [shape=box];
        "Builder subagent implements, tests, commits, self-reviews" [shape=box];
        "Dispatch spec auditor subagent (./spec-reviewer-prompt.md)" [shape=box];
        "Spec auditor confirms code matches spec?" [shape=diamond];
        "Builder subagent fixes spec gaps" [shape=box];
        "Dispatch quality auditor subagent (./code-quality-reviewer-prompt.md)" [shape=box];
        "Quality auditor approves?" [shape=diamond];
        "Builder subagent fixes quality issues" [shape=box];
        "Mark task complete in plan checklist" [shape=box];
    }

    "Read plan, extract all tasks with full text, note context, build task checklist" [shape=box];
    "More tasks remaining?" [shape=diamond];
    "Dispatch final quality auditor for entire implementation" [shape=box];
    "Use godmode:merge-protocol" [shape=box style=filled fillcolor=lightgreen];

    "Read plan, extract all tasks with full text, note context, build task checklist" -> "Dispatch builder subagent (./implementer-prompt.md)";
    "Dispatch builder subagent (./implementer-prompt.md)" -> "Builder subagent has questions?";
    "Builder subagent has questions?" -> "Answer questions, supply context" [label="yes"];
    "Answer questions, supply context" -> "Dispatch builder subagent (./implementer-prompt.md)";
    "Builder subagent has questions?" -> "Builder subagent implements, tests, commits, self-reviews" [label="no"];
    "Builder subagent implements, tests, commits, self-reviews" -> "Dispatch spec auditor subagent (./spec-reviewer-prompt.md)";
    "Dispatch spec auditor subagent (./spec-reviewer-prompt.md)" -> "Spec auditor confirms code matches spec?";
    "Spec auditor confirms code matches spec?" -> "Builder subagent fixes spec gaps" [label="no"];
    "Builder subagent fixes spec gaps" -> "Dispatch spec auditor subagent (./spec-reviewer-prompt.md)" [label="re-audit"];
    "Spec auditor confirms code matches spec?" -> "Dispatch quality auditor subagent (./code-quality-reviewer-prompt.md)" [label="yes"];
    "Dispatch quality auditor subagent (./code-quality-reviewer-prompt.md)" -> "Quality auditor approves?";
    "Quality auditor approves?" -> "Builder subagent fixes quality issues" [label="no"];
    "Builder subagent fixes quality issues" -> "Dispatch quality auditor subagent (./code-quality-reviewer-prompt.md)" [label="re-audit"];
    "Quality auditor approves?" -> "Mark task complete in plan checklist" [label="yes"];
    "Mark task complete in plan checklist" -> "More tasks remaining?";
    "More tasks remaining?" -> "Dispatch builder subagent (./implementer-prompt.md)" [label="yes"];
    "More tasks remaining?" -> "Dispatch final quality auditor for entire implementation" [label="no"];
    "Dispatch final quality auditor for entire implementation" -> "Use godmode:merge-protocol";
}

提示模板

• ./implementer-prompt.md - 派遣构建子 Agent
• ./spec-reviewer-prompt.md - 派遣规格符合性审查员子 Agent
• ./code-quality-reviewer-prompt.md - 派遣代码质量审查员子 Agent

工作流示例

You: I'm using Delegated Execution to implement this plan.

[Read plan file once: docs/plans/feature-plan.md]
[Extract all 5 tasks with full text and context]
[Build task checklist from all tasks]

Task 1: Hook installation script

[Get Task 1 text and context (already extracted)]
[Dispatch builder subagent with full task text + context]

Builder: "Before I start - should the hook install at user level or system level?"

You: "User level (~/.config/godmode/hooks/)"

Builder: "Understood. Implementing now..."
[Later] Builder:
  - Implemented install-hook command
  - Added tests, 5/5 passing
  - Self-review: Noticed I missed --force flag, added it
  - Committed

[Dispatch spec compliance auditor]
Spec auditor: PASS - All requirements met, nothing extraneous

[Get git SHAs, dispatch quality auditor]
Quality auditor: Strengths: Good test coverage, clean code. Issues: None. Approved.

[Mark Task 1 complete]

Task 2: Recovery modes

[Get Task 2 text and context (already extracted)]
[Dispatch builder subagent with full task text + context]

Builder: [No questions, proceeds]
Builder:
  - Added verify/repair modes
  - 8/8 tests passing
  - Self-review: All good
  - Committed

[Dispatch spec compliance auditor]
Spec auditor: FAIL - Issues:
  - Missing: Progress reporting (spec says "report every 100 items")
  - Extra: Added --json flag (not requested)

[Builder fixes issues]
Builder: Removed --json flag, added progress reporting

[Spec auditor re-audits]
Spec auditor: PASS - Spec compliant now

[Dispatch quality auditor]
Quality auditor: Strengths: Solid. Issues (Important): Magic number (100)

[Builder fixes]
Builder: Extracted PROGRESS_INTERVAL constant

[Quality auditor re-audits]
Quality auditor: Approved

[Mark Task 2 complete]

...

[After all tasks]
[Dispatch final quality auditor]
Final auditor: All requirements met, ready to merge

Done!

优势

与手动执行相比：

• 子 Agent 会自然地遵循测试优先
• 每项任务都使用全新的上下文（不会混淆）
• 支持并行安全执行（子 Agent 之间不会相互干扰）
• 子 Agent 可以提问（工作开始前和进行中都可以）

与任务运行器相比：

• 在同一会话中进行（无需交接）
• 持续推进（无需等待）
• 自动设置审查检查点

效率提升：

• 无需额外读取文件（控制器会提供完整文本）
• 控制器会筛选出所需的确切上下文
• 子 Agent 一开始就能收到完整信息
• 在工作开始前提出问题（而不是工作结束后）

质量门禁：

• 自我审查可在交接前发现问题
• 两阶段审查：先检查规格符合性，再检查代码质量
• 审查循环确保修复确实有效
• 规格符合性检查可避免构建过多或不足
• 代码质量检查可确保实现精良

成本：

• 调用更多子 Agent（每项任务需要构建者 ＋ 2 名审查员）
• 控制器需要做更多准备工作（预先提取所有任务）
• 审查循环会增加迭代次数
• 但能及早发现问题（比之后调试更省成本）

防护措施

绝不：

• 未经用户明确同意，就在 main/master 分支上开始实现
• 跳过审查（规格符合性或代码质量）
• 在问题尚未解决时继续推进
• 并行派遣多个构建子 Agent（会产生冲突）
• 让子 Agent 阅读计划文件（应直接提供完整文本）
• 跳过背景说明（子 Agent 需要了解任务所处的位置）
• 忽略子 Agent 的问题（让其继续之前先作答）
• 在规格符合性方面接受“差不多”（审查员发现问题就意味着尚未完成）
• 跳过审查循环（审查员发现问题 → 构建者修复 → 再次审查）
• 以构建者的自我审查替代实际审查（两者都不可少）
• 在规格符合性通过之前开始质量审查（顺序错误）
• 在任一审查仍有未解决问题时转到下一项任务

如果子 Agent 提问：

• 清晰、完整地作答
• 视需要补充背景信息
• 不要催促其开始实现

如果审查员发现问题：

• 由构建者（同一子 Agent）修复
• 审查员再次审查
• 重复此过程，直到通过
• 不要跳过重新审查

如果子 Agent 未能完成任务：

• 派遣一名修复子 Agent，并提供具体说明
• 不要手动修复（以免造成上下文污染）

关联

必需的工作流技能：

• godmode:workspace-isolation - 必需：开始前设置隔离工作区
• godmode:task-planning - 制定此技能所执行的计划
• godmode:quality-gate - 提供供审查员子 Agent 使用的代码审查模板
• godmode:merge-protocol - 所有任务完成后结束开发

子 Agent 应使用：

• godmode:test-first - 子 Agent 为每项任务遵循测试优先

替代工作流：

• godmode:task-runner - 使用独立会话，而非在同一会话中执行