playbooks

home / skills / bahayonghang / my-claude-code-settings / interview-plan

interview-plan skill

/content/skills/workflow-skills/interview-plan

This skill conducts Socratic interviews to refine requirements into executable plans and directly invokes Plan mode for actionable outcomes.

npx playbooks add skill bahayonghang/my-claude-code-settings --skill interview-plan

Review the files below or copy the command above to add this skill to your agents.

Files (5)
SKILL.md
2.3 KB
---
name: interview-plan
description: 通过苏格拉底式访谈细化需求，直接调用原生 Plan 模式生成可执行计划。适用于需求细化、技术方案设计等场景。
version: 3.0
argument-hint: [draft-or-requirements-path]
allowed-tools: Read, Glob, Grep, AskUserQuestion, EnterPlanMode
category: workflow
tags: [interview, planning, requirements, design, architecture]
---

# 访谈式计划

通过苏格拉底式访谈将模糊需求转化为清晰的工程规格，然后直接调用平台原生 Plan 模式生成可执行的实施计划。

## 步骤

1. 若 `$ARGUMENTS` 为空，提示用户提供需求文档路径（默认查找 `plan.md`）。若文件不存在，报错退出。
2. 读取目标文件，分析核心业务目标、方案完整度、模糊/矛盾/缺失点及潜在风险。
3. 读取 `resources/INTERVIEW_PRINCIPLES.md`，按其提问原则进行苏格拉底式访谈。
4. 读取 `resources/INTERVIEW_DIMENSIONS.md`，按 A-G 维度框架逐项向用户提问：
   - 必须覆盖 A（工程原则审查），其余按项目类型选择性覆盖。
   - **必须使用 `AskUserQuestion` 工具**呈现每轮问题，将 2-3 个收敛性选项映射为可点击的交互式选项卡（`options`），禁止以纯文本 A/B/C 形式输出。
   - 每个 `AskUserQuestion` 调用包含 1-4 个问题（`questions`），每个问题 2-4 个选项（`options`），用户可通过 UI 直接点选。
   - 相互独立的问题应合并到同一次 `AskUserQuestion` 调用中，减少交互轮次。
5. 若涉及高风险操作，读取 `resources/RISK_PROTECTION.md` 并发出告警，等待用户明确授权。
6. 访谈完成后，在内部整理访谈结论摘要（业务目标、架构决策、技术约束、风险评估等）。
7. **直接调用 `EnterPlanMode` 进入原生 Plan 模式**，在计划文件中写入访谈结论与实施计划：
   - 任务分解与优先级排序
   - 依赖关系与执行顺序
   - 每个任务的验收标准
   - 风险缓解措施
8. 用户审批计划后，调用 `ExitPlanMode` 进入执行阶段。

## 技术约束

- 路径使用双引号包裹，正斜杠 `/` 分隔
- 优先使用 `rg` 替代 `grep`
- **禁止自动执行** `commit`、`push` 等 Git 写操作
- 遵循 Read-before-Write 原则
- **不生成独立的 spec.md 文件**，所有规格内容直接融入原生 Plan

Overview

This skill conducts Socratic interviews to turn vague requirements into clear engineering specifications and then invokes the platform's native Plan mode to produce an executable implementation plan. It reads a target document, probes gaps through structured questions, and writes the final plan directly into Plan mode for review and approval. The flow enforces safety checks and respects repository write constraints.

How this skill works

The skill starts by locating a requirement file (defaults to "plan.md") and analyzing business goals, completeness, ambiguities, and risks. It loads interview principles and dimension frameworks from "resources/INTERVIEW_PRINCIPLES.md" and "resources/INTERVIEW_DIMENSIONS.md" and runs a Socratic Q&A using the AskUserQuestion tool. After summarizing conclusions and risk mitigations, it calls EnterPlanMode to create the plan (tasks, dependencies, acceptance criteria, priorities, risk mitigations) and waits for user approval before calling ExitPlanMode to proceed.

When to use it

  • Refining a vague product requirement into an actionable engineering plan
  • Designing technical solutions where constraints or tradeoffs are unclear
  • Preparing detailed implementation plans for handoff to engineers
  • Auditing existing proposals for missing assumptions, risks, or conflicts
  • Scoping projects that require prioritized task breakdowns and acceptance criteria

Best practices

  • Provide the target requirement file path or keep "plan.md" in repo root for fast start
  • Allow the tool to run multiple concise AskUserQuestion rounds rather than long free-text sessions
  • Keep interview answers committed in the UI; avoid embedding large attachments in questions
  • Follow Read-before-Write: inspect files with "rg" and quoted paths (e.g. "resources/INTERVIEW_DIMENSIONS.md") before modifying Plan
  • Do not expect automatic Git commit/push; all Git write actions are blocked by design

Example use cases

  • You have a one-page product brief and need a prioritized engineering plan with acceptance criteria
  • A design doc has ambiguous performance targets; run targeted interviews to resolve tradeoffs and create implementation tasks
  • Risk-heavy operations (infra changes, migrations) where the skill must surface hazards and request explicit authorization
  • Converting stakeholder interviews into a sequenced task queue with dependencies for Sprints

FAQ

What file will the skill read by default?

If no path is provided it looks for "plan.md" in the repo root; if the file is missing it reports an error and stops.

How are interview prompts delivered?

Questions are delivered using the AskUserQuestion tool. Each call contains 1–4 questions with 2–4 clickable options, grouped to minimize rounds.

Will this skill push commits or modify Git history?

No. The skill cannot perform automatic Git write operations such as commit or push. Plan content is written only within the native Plan mode and Git actions remain manual.