implementing

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Implementing

计划执行

You are implementing an approved technical plan, executing it phase by phase with verification at each step. Each phase is executed as a background sub-agent via

desplega:phase-running

— the main session acts as an orchestrator.

你正在执行一份已获批的技术计划，分阶段逐步推进，每一步都包含验证环节。每个阶段通过

desplega:phase-running

由后台子Agent执行——主会话担任编排者角色。

Working Agreement

工作约定

All user-facing questions go through
AskUserQuestion
— see

desplega:ask-user

for conventions. Never ask in chat as plain bullets.

All read/research/validation work goes through sub-agents — keep raw tool output out of the main session. Default to

run_in_background: true

The autonomy mode (below) controls how often you check in. AskUserQuestion is always the mechanism.

File-review is on by default — when significant changes land in a phase, invoke

/file-review:file-review <path>

for inline feedback (skip only if Autopilot).

所有面向用户的问题需通过
AskUserQuestion
发起——相关规范请参考

desplega:ask-user

。禁止在聊天中以普通项目符号形式提问。

所有读取/调研/验证工作需通过子Agent完成——请勿将原始工具输出直接展示在主会话中。默认设置

run_in_background: true

。

自主模式（如下）将控制你同步进度的频率。所有用户交互必须通过AskUserQuestion机制完成。

文件审核默认开启——当某一阶段产生重大变更时，调用

/file-review:file-review <path>

获取内联反馈（仅Autopilot模式下可跳过）。

When to Use

适用场景

This skill activates when:

User invokes
```
/implement-plan
```
command

Another skill references

**REQUIRED SUB-SKILL:** Use desplega:implementing

User asks to execute or implement an existing plan

本技能在以下场景激活：

用户调用
```
/implement-plan
```
命令

其他技能引用

**REQUIRED SUB-SKILL:** Use desplega:implementing

用户要求执行或落地现有计划

Autonomy Mode

自主模式

Adapt your behavior based on the autonomy mode:

Mode	Behavior
Autopilot	Execute all phases, pause only for manual verification or blockers
Critical (Default)	Pause between phases for approval, ask when mismatches found
Verbose	Check in frequently, confirm before each major change

The autonomy mode is passed by the invoking command. If not specified, default to Critical.

根据自主模式调整行为：

模式	行为
Autopilot（自动驾驶）	执行所有阶段，仅在需要手动验证或遇到阻塞时暂停
Critical（关键校验，默认）	阶段间暂停以获取批准，发现不符情况时发起询问
Verbose（详细交互）	频繁同步进度，每次重大变更前确认

自主模式由调用命令传入。若未指定，默认使用Critical模式。

Initial Setup Questions

初始设置问题

After establishing user preferences, use AskUserQuestion to gather implementation-specific details:

确认用户偏好后，使用AskUserQuestion收集执行相关细节：

1. Branch/Worktree Setup

1. 分支/工作树设置

First, check the current branch:

git branch --show-current

Then check if the

wts

plugin is available (look for

wts:wts

in available skills).

If wts plugin is installed, use AskUserQuestion with these options:

Question	Options
"You're currently on branch `<current-branch>` . Where would you like to implement?"	1. Continue on current branch, 2. Create a new branch, 3. Create a wts worktree

If wts plugin is NOT installed, use AskUserQuestion with:

Question	Options
"You're currently on branch `<current-branch>` . Where would you like to implement?"	1. Continue on current branch, 2. Create a new branch

首先，检查当前分支：

git branch --show-current

然后检查

wts

插件是否可用（在可用技能中查找

wts:wts

）。

若已安装wts插件，使用AskUserQuestion提供以下选项：

问题	选项
"你当前位于分支 `<current-branch>` 。希望在哪里执行计划？"	1. 继续在当前分支执行，2. 创建新分支，3. 创建wts工作树

若未安装wts插件，使用AskUserQuestion提供以下选项：

问题	选项
"你当前位于分支 `<current-branch>` 。希望在哪里执行计划？"	1. 继续在当前分支执行，2. 创建新分支

2. Commit Strategy

2. 提交策略

Use AskUserQuestion with these options:

Question	Options
"How would you like to handle commits during implementation?"	1. Commit after each phase (Recommended for complex plans), 2. Commit at the end (Single commit for all changes), 3. Let me decide as I go

If "Commit after each phase" is selected:

After completing each phase's verification, create a commit with message:
```
[Phase N] <phase name>
```
Use the plan's phase descriptions for commit messages

Store these preferences and apply them throughout the implementation.

使用AskUserQuestion提供以下选项：

问题	选项
"执行过程中希望如何处理代码提交？"	1. 每个阶段完成后提交（复杂计划推荐），2. 全部完成后一次性提交，3. 由我根据情况决定

若选择“每个阶段完成后提交”：

完成每个阶段的验证后，创建提交，信息为：
```
[Phase N] <phase name>
```
使用计划中的阶段描述作为提交信息

保存这些偏好并在整个执行过程中遵循。

Prior Learning Recall

过往学习回顾

OPTIONAL SUB-SKILL: If

~/.agentic-learnings.json

exists, run

/learning recall <current topic>

to check for relevant prior learnings before proceeding.

可选子技能： 若

~/.agentic-learnings.json

文件存在，在开始前运行

/learning recall <current topic>

查看相关过往学习内容。

Getting Started

开始执行

When given a plan path:

Read the plan completely
Check for existing checkmarks (
```
- [x]
```
)
Set plan status to
```
status: in-progress
```
by editing the frontmatter
```
status
```
field. This signals to progress-tracking hooks which plan is active.
Read files fully - never use limit/offset parameters
Think deeply about how the pieces fit together
Create a todo list to track progress
Start implementing if you understand what needs to be done

If no plan path provided, ask for one.

当收到计划路径时：

完整阅读计划内容
检查已完成的标记（
```
- [x]
```
）
编辑计划前置元数据的
```
status
```
字段，将其设置为
```
status: in-progress
```
。这会向进度跟踪钩子发送信号，标识当前活跃的计划。
完整读取文件——切勿使用limit/offset参数
深入思考各部分的关联方式
创建待办列表跟踪进度
若理解需求则开始执行

若未提供计划路径，请向用户索取。

Implementation Philosophy

执行理念

Plans are carefully designed, but reality can be messy. Your job is to:

Follow the plan's intent while adapting to what you find
Implement each phase fully before moving to the next
Verify your work makes sense in the broader codebase context
Update checkboxes in the plan as you complete sections

When things don't match the plan exactly, think about why and communicate clearly.

计划经过精心设计，但实际情况可能复杂多变。你的职责是：

遵循计划意图，同时根据实际情况灵活调整
完全完成一个阶段后再推进到下一个阶段
验证工作内容在整个代码库语境中的合理性
完成计划中的章节后更新复选框状态

当实际情况与计划不完全匹配时，思考原因并清晰沟通。

Handling Mismatches

处理不符情况

If you encounter a mismatch (and autonomy mode is not Autopilot):

STOP and think deeply about why the plan can't be followed
Use AskUserQuestion to present the issue and get direction:

Question	Options
"Issue in Phase [N]: Expected [what the plan says], Found [actual situation]. Why this matters: [explanation]. How should I proceed?"	1. Adapt plan to match reality, 2. Proceed as originally planned, 3. Stop and discuss

In Autopilot mode, use best judgment and document decisions in comments.

若遇到不符情况（且自主模式不是Autopilot）：

停止操作并深入思考计划无法执行的原因
使用AskUserQuestion呈现问题并获取指导：

问题	选项
"第[N]阶段问题：计划预期[计划内容]，实际情况[实际情况]。影响说明：[解释]。应如何推进？"	1. 调整计划以匹配实际情况，2. 按原计划执行，3. 停止并讨论

在Autopilot模式下，自行判断并在注释中记录决策。

Verification Approach

验证流程

Each phase is executed via a background sub-agent running

desplega:phase-running

Read the phase overview in the main session (minimal context usage)
Spawn a
desplega:phase-running
agent in background with plan path + phase number:
- Use the Agent tool with
```
run_in_background: true
```
- Pass the plan path and phase number as context
Wait for agent completion — you'll be notified when it finishes
Review the agent's report — check status (completed/blocked/failed), changed files, verification results
Check QA Doc status — if the phase agent reports
```
QA Doc: <path>
```
, the linked QA doc has scenarios that need execution. Invoke
```
desplega:qa
```
against the QA doc path. (For
```
QA: n/a
```
, proceed normally.) Note: Automated QA items inside the phase's Success Criteria block are already handled by the phase agent — only the linked QA doc needs separate orchestration here.
Handle manual verification with the user — present the manual verification items from the phase
Proceed to next phase after user confirms

The implementing skill is an orchestrator — it coordinates phases, handles human checkpoints, and manages cross-phase decisions, but delegates actual implementation work to phase-runner sub-agents.

Executor routing: if the

desplega:delegate-work

skill is available, pick each phase's executor per its routing matrix instead of defaulting to a

phase-running

sub-agent — a phase may route to a Codex variant or a specific Claude model tier. Only the executor choice changes; everything else here (autonomy modes, checkpoints, plan bookkeeping, verification, commit strategy) stays unchanged.

每个阶段由运行

desplega:phase-running

的后台子Agent执行：

在主会话中阅读阶段概述（最小化上下文使用）
在后台启动
desplega:phase-running
Agent，传入计划路径和阶段编号：
- 使用Agent工具并设置
```
run_in_background: true
```
- 将计划路径和阶段编号作为上下文传递
等待Agent完成——完成后会收到通知
审核Agent报告——检查状态（已完成/阻塞/失败）、变更文件、验证结果
检查QA文档状态——若阶段Agent报告
```
QA Doc: <path>
```
，则关联的QA文档包含需要执行的场景。针对该QA文档路径调用
```
desplega:qa
```
。（若为
```
QA: n/a
```
，则正常推进。）注意：阶段成功标准块内的自动化QA项已由阶段Agent处理——仅需单独编排关联的QA文档。
与用户进行手动验证——展示阶段中的手动验证项
用户确认后推进到下一阶段

执行技能是编排者——负责协调阶段、处理人工检查点、管理跨阶段决策，但将实际执行工作委托给阶段执行子Agent。

执行器路由：若

desplega:delegate-work

技能可用，根据路由矩阵为每个阶段选择执行器，而非默认使用

phase-running

子Agent——某一阶段可能路由到Codex变体或特定Claude模型层级。仅执行器选择会改变；其他规则（自主模式、检查点、计划记录、验证、提交策略）保持不变。

Pause for Human Verification (if not Autopilot or executing multiple phases)

暂停等待人工验证（非Autopilot模式或非连续执行多阶段时）

Phase [N] Complete - Ready for Manual Verification

Automated verification passed:
- [List automated checks that passed]

Please perform the manual verification steps listed in the plan:
- [List manual verification items from the plan]

Let me know when manual testing is complete so I can proceed to Phase [N+1].

If instructed to execute multiple phases consecutively, skip the pause until the last phase.

Do not check off manual testing items until confirmed by the user.

Phase [N] Complete - Ready for Manual Verification

Automated verification passed:
- [List automated checks that passed]

Please perform the manual verification steps listed in the plan:
- [List manual verification items from the plan]

Let me know when manual testing is complete so I can proceed to Phase [N+1].

若被指示连续执行多个阶段，仅在最后一个阶段后暂停。

除非用户确认完成，否则请勿标记手动测试项为已完成。

If You Get Stuck

遇到阻塞时

When something isn't working as expected:

Make sure you've read and understood all relevant code
Consider if the codebase has evolved since the plan was written
Present the mismatch clearly and ask for guidance (unless Autopilot)
Use sub-tasks sparingly for targeted debugging

当出现预期外问题时：

确保已阅读并理解所有相关代码
考虑代码库自计划编写以来是否有更新
清晰呈现不符情况并请求指导（Autopilot模式除外）
谨慎使用子任务进行针对性调试

Resuming Work

恢复工作

If the plan has existing checkmarks:

Trust that completed work is done
Pick up from the first unchecked item
Verify previous work only if something seems off

若计划已有完成标记：

信任已完成的工作
从第一个未完成的项开始
仅当发现异常时才验证之前的工作

Learning Capture

学习内容捕获

OPTIONAL SUB-SKILL: If significant insights, patterns, gotchas, or decisions emerged during this workflow, consider using

desplega:learning

to capture them via

/learning capture

. Focus on learnings that would help someone else in a future session.

可选子技能： 若在工作流程中获得重要见解、模式、注意事项或决策，可考虑使用

desplega:learning

通过

/learning capture

捕获这些内容。重点记录能帮助未来会话中其他用户的学习内容。

Completing Implementation

完成执行

When all phases are complete and verified:

Set the plan frontmatter
```
status
```
field to
```
completed
```
.
Ensure automated verification checkboxes are updated.
Leave manual verification checkboxes unchecked until the user confirms completion.
Offer post-implementation auditing: "Would you like me to run
```
/verify-plan
```
for a post-implementation audit, then
```
/review
```
for a final quality check?"

Remember: You're implementing a solution, not just checking boxes. Keep the end goal in mind.

所有阶段完成并验证后：

将计划前置元数据的
```
status
```
字段设置为
```
completed
```
。
确保自动化验证复选框已更新。
手动验证复选框保持未勾选状态，直到用户确认完成。
提供执行后审计选项："是否需要我运行
```
/verify-plan
```
进行执行后审计，再运行
```
/review
```
进行最终质量检查？"

请记住：你是在落地解决方案，而非仅仅勾选任务框。时刻牢记最终目标。

Review Integration

审核集成

File-review is on by default (unless Autopilot):

After significant code changes in each phase, invoke
```
/file-review:file-review <changed-file-path>
```
for inline human comments
Process feedback with the
```
file-review:process-review
```
skill before moving to the next phase

文件审核默认开启（Autopilot模式除外）：

每个阶段产生重大代码变更后，调用
```
/file-review:file-review <changed-file-path>
```
获取人工内联反馈
在推进到下一阶段前，使用
```
file-review:process-review
```
技能处理反馈