360 Secure SkillHub
360 安全技能中心

Planning Superpowers Pro

通过 task_plan.md、findings.md、progress.md 三个 Markdown 文件把计划、调研与日志持久化到磁盘，强制先规划、先写失败测试再实现，并规定系统化调试步骤与子代理委派模式；附带初始化脚本、会话恢复脚本及模板与参考文档。

jaccchina-ai
jaccchina-ai
热度 20.11
最近更新 2026-03-31 17:11:52
分类：开发依赖：代码执行 / 读写文件 / 内部调用大模型应用场景：代码生成 / 调试排错 / 任务管理 / 工作流 ...

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add planning-superpowers-pro
Metadata
name
planning-superpowers-pro
description
Advanced task planning and developer toolkit. Integrates TDD (Test-Driven Development) workflows, systematic debugging protocols, and file-driven task tracking to enhance execution quality and reliability for complex engineering tasks.

Planning Superpowers Pro

Combined file-based planning (Planning with Files) and spec-first TDD development (Superpowers).

The Core Mandate

  1. Plan First: Never start a task without task_plan.md.
  2. TDD Required: No production code without a failing test first.
  3. Systematic Debugging: No fixes without root cause investigation.
  4. Persistent Memory: Use markdown files (task_plan.md, findings.md, progress.md) as your "working memory on disk."

1. Initialization (Start of Task)

Before doing anything else, initialize the planning files.

Recovery Check

If task_plan.md already exists, read it along with progress.md and findings.md to restore context. Run catchup:

python3 scripts/session-catchup.py "$(pwd)"

New Project Setup

Create the holy trinity of planning:

  • task_plan.md: The roadmap and state.
  • findings.md: Research, decisions, and discovered data.
  • progress.md: Detailed session log and test results.

Use scripts/init-session.sh to scaffold these.

2. Planning Phase (TDD Planning)

Every task must be broken down into "TDD-sized" chunks (2-10 minutes each). A valid task in task_plan.md must follow this micro-loop:

  1. Write a failing test.
  2. Watch it fail.
  3. Write minimal code to pass.
  4. Watch it pass.
  5. Refactor.
  6. Commit.

Hard Gate: Do not write production code until a failing test is verified in progress.md.

3. Execution & Subagents

For complex tasks, use sessions_spawn to delegate to subagents.

Implementer Subagent Prompt Pattern:

Goal: [Target behavior]
Context: [Link to task_plan.md / design docs]
TDD Requirement: Write test first -> watch fail -> implement -> watch pass.
Verification: Run [test command].

4. Systematic Debugging

When a test fails or an error occurs, do not "just try a fix." Follow the Systematic Debugging Protocol:

  1. Investigate: Read full stack traces. Trace data flow to the source.
  2. Reproduce: Ensure the failure is consistent.
  3. Hypothesize: State "I think [cause] because [evidence]" before editing.
  4. Test Hypothesis: Prove/disprove the hypothesis with a minimal test.
  5. Fix at Root: Fix the cause, not the symptom. Log in task_plan.md.

5. File Management Rules

File Update Frequency Content
task_plan.md After each phase Status, Decisions, Error Log
findings.md Every 2 tool calls Research, API keys, URLs, Data
progress.md Every action Test outputs, CLI logs, small wins

The 2-Action Rule

After every 2 search or browser operations, move data from context to findings.md. Multimodal data (images/PDFs) must be summarized as text immediately.

6. Finishing

  1. Verify all tests pass.
  2. Update task_plan.md to complete.
  3. Provide final summary and clean up temporary files.

Templates & References

  • Templates: templates/task_plan.md, templates/findings.md, templates/progress.md
  • Reference Docs: references/tdd.md, references/systematic-debugging.md, references/brainstorming.md