SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsexecuting-plans

executing-plans

Agents & models

View repository

Updated 10-16 09:15

About this skill

Executing Plans

Overview

Load the plan, review it critically, execute all tasks, and report when done.

At the start, announce: "I am using the executing-plans skill to implement this plan."

Note: Tell your human partner that Superpowers works much better with subagent support. If running on a platform that supports subagents (such as Claude Code or Codex), the quality of work will be significantly higher. If subagents are available, use superpowers:subagent-driven-development instead of this skill.

Process

Step 1: Load and review the plan

  1. Read the plan file
  2. Review it critically — identify any issues or concerns in the plan
  3. If there are concerns: raise them with your human partner before starting
  4. If there are no concerns: create TodoWrite and continue

When reviewing, focus on checking:

  • Are there missing dependencies between steps? (A depends on B, but B is ordered after A)
  • Are the verification conditions clear? ("Confirm it works" does not count; "run npm test all passing" counts)
  • Are there implicit environment assumptions? (Node version, database connection, API Key)

Review example:

计划文件：docs/plan.md
任务清单：5 个任务

审查发现：
- 任务 3（添加数据库迁移）应在任务 2（编写数据模型）之后，顺序正确 ✓
- 任务 4 的验证条件写的是"确认功能正常"→ 需澄清：具体跑什么测试？
- 计划未提及 Python 版本要求 → 需确认

向伙伴提出：
"计划整体可执行。有两个问题：(1) 任务 4 的验证条件不够具体，建议改为
'运行 pytest tests/test_api.py 全部通过'；(2) 需要确认 Python 版本要求。"

Step 2: Execute tasks

For each task:

  1. Mark as in progress — update TodoWrite
  2. Understand the goal — reread the task description and clarify the completion criteria
  3. Execute the implementation — execute strictly according to the plan steps (the plan already has small steps)
  4. Run verification — run tests or checks as required
  5. Commit changes — commit once after each completed task, with the commit message referencing the task number
  6. Mark as completed — update TodoWrite

Rhythm for each task:

--- 任务 2/5：添加用户验证 ---
[标记进行中]

目标：为 /api/users 添加输入验证
完成标准：所有验证测试通过，无效输入返回 400

[实现]
- 添加 validateUser() 中间件
- 编写 3 个验证规则（email 格式、密码强度、用户名长度）

[验证]
$ npm test -- --grep "validation"
  ✓ 拒绝无效 email (12ms)
  ✓ 拒绝弱密码 (8ms)
  ✓ 拒绝过长用户名 (5ms)
  3 passing

[提交]
$ git add src/middleware/validate.js tests/validation.test.js
$ git commit -m "feat: 添加用户输入验证（任务 2/5）"

[标记完成]
--- 任务 2/5 完成 ---

Batch review checkpoints:

  • After every 3 completed tasks, pause and review: is the overall direction still right? Has there been any deviation from the plan?
  • If you find a problem with earlier implementation, fix it first before continuing; do not keep going with the problem

Step 3: Handle common exceptions

Test failure:

  1. Read the error message and locate the cause of the failure
  2. Distinguish: is it an implementation bug? Or is the test itself problematic? Or is the plan description wrong?
  3. Implementation bug → fix and rerun
  4. Test problem → fix the test, explain to your partner
  5. Plan error → stop, report to your partner, and suggest a correction

Missing dependency:

任务 3 需要 Redis 连接，但计划中没有提及 Redis 配置。
→ 停止执行
→ 向伙伴报告："任务 3 需要 Redis，计划中未包含配置步骤。
   建议：在任务 3 前插入 '配置 Redis 连接' 步骤。"

Unclear instructions:

  • Do not guess intent, do not "reasonably infer"
  • List your understanding and confusion, and let your partner clarify
  • Wait for a reply before continuing

Step 4: Finish development

After all tasks are complete and verified:

  • Announce: "I am using the finishing-a-development-branch skill to finish this work."
  • Required sub-skill: Use superpowers:finishing-a-development-branch
  • Follow that skill's guidance to verify tests, present options, and execute the choice

Completion report template:

## 执行报告

**计划：** docs/plan.md
**分支：** feature/user-validation
**任务：** 5/5 已完成

### 完成的任务
1. ✅ 初始化项目结构
2. ✅ 添加用户验证
3. ✅ 添加数据库迁移
4. ✅ 实现 API 端点
5. ✅ 添加集成测试

### 验证结果
- 单元测试：23/23 通过
- 集成测试：8/8 通过
- lint 检查：0 个警告

### 偏离计划的地方
- 任务 3：Redis 配置从 env 改为 config.yaml（经伙伴同意）

### 下一步
按 finishing-a-development-branch 技能处理合并/PR

When to stop and ask for help

Stop execution immediately in the following situations:

  • Encountering a blocker (missing dependency, failing test, unclear instructions)
  • The plan has serious flaws that prevent starting
  • You do not understand an instruction
  • Verification repeatedly fails (the same test fails more than 2 times)

When unsure, ask; do not guess.

When to return to a previous step

Return to review (step 1) when:

  • The partner updated the plan based on your feedback
  • The fundamental approach needs to be reconsidered

Do not force your way through blockers — stop and ask.

Notes

  • Critically review the plan first
  • Execute strictly according to the plan steps
  • Do not skip verification
  • Commit each task separately, and reference the task number in the commit message
  • Reference the corresponding skill when the plan requires it
  • Stop when encountering a blocker; do not guess
  • Never start implementation on the main/master branch without the user's explicit consent

Integration

Required workflow skills:

  • superpowers:using-git-worktrees - Required: establish an isolated workspace before starting
  • superpowers:writing-plans - Create the plan that this skill will execute
  • superpowers:finishing-a-development-branch - Wrap up development after all tasks are completed