跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
搜索与研究

plan-feature

Pre-implementation planning — interview to clarify scope, then compile into a PRD. Triggers "help me figure out", "vague scope", "define requirements" (discovery phase); "PRD", "requirements document", "product spec", "feature spec", "write requirements" (PRD phase).

darkroomengineering · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add darkroomengineering/cc-settings --skill plan-feature

技能说明


name: plan-feature description: Pre-implementation planning — interview to clarify scope, then compile into a PRD. Triggers "help me figure out", "vague scope", "define requirements" (discovery phase); "PRD", "requirements document", "product spec", "feature spec", "write requirements" (PRD phase). context: fork agent: planner

Plan Feature

Two-phase pre-implementation planning: clarify requirements via interview, then compile a complete PRD.

Phase 1: Discovery

Help clarify requirements and scope through structured questioning.

Interview Framework

The interview is a fog-of-war walk across four quadrants of the unknown. Open by listing the known knowns (what the user has already decided), then work the questions below to surface known unknowns (open questions they're aware of), unknown knowns (constraints they hold but haven't said — the questions in 2–4 exist to shake these loose), and close by hunting unknown unknowns ("what would surprise us mid-build? what reference implementation should we read first?"). A discovery that ends with all four quadrants visited produces a PRD that doesn't get re-planned in week two.

Goal Quality Bar (gate before interviewing)

Before opening the interview questions below, try to state the goal in one line that answers five things:

  • What's true when this is done?
  • What evidence shows it — a command, a test, a metric, a reviewed artifact?
  • What threshold counts as success — pass/fail, or a number?
  • What's explicitly out of scope, where that would matter?
  • What's the stop condition — the point where you ask the user instead of guessing?

If a clean one-liner falls out, skip straight to Phase 2's clarifying questions — the interview below exists for when it doesn't.

Reject pure activity goals — "make progress," "keep investigating," "improve things" — until sharpened into a verifiable outcome:

  • Weak: "Make checkout faster." Sharpened: "Reduce checkout API p95 latency below 250ms for the documented slow path, verified with npm run test:checkout and 3 consecutive benchmark runs under 250ms."
  • Weak: "Clean up the auth code." Sharpened: "Resolve the open change-request threads on PR 123, touching only the affected auth files and tests, verified with the targeted auth test command plus gh pr view 123 showing no unresolved threads."

Pick the validator shape by domain:

DomainSuccess looks like
BugReproduce first, fix second — a failing-then-passing test or repro script
TestThe exact command and its pass condition
PerformanceMetric + threshold + measurement method + run count
QualityReviewed examples, or lint/typecheck/test passing
ResearchThe decision the research needs to unblock
OpsHealthy state + monitoring window + rollback trigger

Goal Quality Bar adapted from openai/skills define-goal, Apache-2.0.

1. Understand the Goal

  • What problem are you solving?
  • Who is this for?
  • What does success look like?

2. Define Scope

  • What must be included (MVP)?
  • What's nice to have (future)?
  • What's explicitly out of scope?

3. Identify Constraints

  • Timeline constraints?
  • Technical constraints?
  • Resource constraints?

4. Clarify Details

  • What are the inputs/outputs?
  • What are the edge cases?
  • What are the error scenarios?

5. Validate Understanding

  • Summarize back what you heard
  • Confirm priorities
  • Identify open questions

Output

## Discovery Summary: [Feature/Project]

### Goal
[Clear statement of what we're building and why]

### Requirements
**Must Have (MVP)**
- [ ] Requirement 1
- [ ] Requirement 2

**Nice to Have**
- [ ] Feature A
- [ ] Feature B

**Out of Scope**
- Not doing X
- Not doing Y

### Technical Approach
[High-level approach]

### Open Questions
- [ ] Need to clarify: ...
- [ ] Decision needed: ...

### Next Steps
1. [First action]
2. [Second action]

Remember

  • Ask, don't assume
  • Summarize frequently
  • Document decisions
  • Store requirements as learnings

Phase 2: PRD Compilation

Structured 6-phase process to produce a complete PRD from a feature idea, including user stories, task breakdown, and parallel execution plan.

Workflow

Phase 1: Clarifying Questions

Ask 5-8 targeted questions to fill gaps. Use smart defaults so the user can skip.

## Clarifying Questions

1. **Target users?** [default: existing app users]
2. **Platform scope?** [default: web only]
3. **Auth required?** [default: yes, existing auth]
4. **Performance targets?** [default: <2.5s LCP, <200ms INP]
5. **Accessibility level?** [default: WCAG 2.1 AA]
6. **Data persistence?** [default: existing database]
7. **Mobile responsive?** [default: yes]
8. **Analytics needed?** [default: basic events]

Press enter to accept all defaults, or answer specific questions.

Phase 2: Scope Definition

Define what is IN and OUT of scope.

## Scope

### In Scope
- [feature 1]
- [feature 2]

### Out of Scope
- [explicitly excluded 1]
- [explicitly excluded 2]

### Assumptions
- [assumption 1]
- [assumption 2]

Phase 3: User Stories

Write user stories with acceptance criteria.

## User Stories

### US-1: [Title]
**As a** [role]
**I want** [capability]
**So that** [benefit]

**Acceptance Criteria:**
- [ ] Given [context], when [action], then [result]
- [ ] Given [context], when [action], then [result]

**Priority:** P1 | P2 | P3
**Complexity:** trivial | small | medium | large | epic

Phase 4: Task Breakdown

Break into implementable tasks with metadata. Dependencies/Blocks feed the batch algorithm and must match the Functional DAG's joins from Phase 5 — on conflict the DAG wins, so reconcile the metadata to it rather than redrawing to match a stale field.

## Tasks

### T-1: [Title]
- **Description:** [what to implement]
- **User Story:** US-1
- **Priority:** 1 (1=highest, 5=lowest)
- **Complexity:** medium
- **Estimated Tokens:** 15
阅读完整来源 ↗