AgentSkillsCN

workflow-discuss

当 /work 命令启动新工作区时使用。引导讨论环节——逐个提出澄清性问题,在 DISCUSSION.md 中记录决策要点,并在 ASSUMPTIONS.md 中追踪各项假设。

SKILL.md
--- frontmatter
name: workflow-discuss
description: Use when the /work command starts a new workspace. Guides the discussion phase — asks clarifying questions one at a time, captures decisions in DISCUSSION.md, tracks assumptions in ASSUMPTIONS.md.

Workflow: Discussion Phase

Guide the user through a structured discussion to understand what they want to build/fix/refactor. Capture everything in DISCUSSION.md.

Inputs

  • Workspace path: .working/{type}/{slug}/ (provided by /work command)
  • User's description: The original intent (from /work arguments)
  • STATUS.md: Already created by /work command with type, slug, timestamps

Skills Integration

Before starting the discussion, invoke these skills to inform your exploration:

Always invoke

  • brainstorming (superpowers) — Use this to explore the solution space before narrowing scope. Present 2-3 creative approaches to the user during discussion.

Invoke based on work type

Project skills (from .claude/skills/):

  • page-building — If the work involves creating or modifying pages, invoke to understand the page architecture patterns (listing vs. detail vs. browse).
  • component-architecture — If the work involves new or restructured components, invoke to understand file organization, composition, and prop design patterns.
  • data-fetching — If the work involves API data, invoke to understand the query pipeline (API client → function → query options → Query wrapper).
  • styling-design — If the work involves visual changes, invoke to understand the color system, animation patterns, and responsive approach.
  • query-autonomy — If the work involves refactoring prop passthrough into independent queries.

Plugin skills/tools:

  • frontend-design (plugin) — If the work involves building new UI with significant visual design. Use to inform bold aesthetic direction and avoid generic AI aesthetics.
  • feature-dev:code-explorer (agent) — Spawn this agent to deeply analyze existing features before discussing changes. It traces execution paths and maps architecture layers.
  • context7 (MCP tools) — CRITICAL: Never assume library APIs. Use resolve-library-id + query-docs to look up current documentation for Next.js, TanStack Query, Radix/shadcn, nuqs, or any library used in this project. Do this during discussion when API behavior matters for a decision.

Reference the relevant skills when asking questions — e.g., "Based on our page-building patterns, this would follow the Detail Page structure (Header + Stats + Tabs). Does that match your intent?"

Process

1. Understand the intent

Read the user's description and ask clarifying questions one at a time. Prefer multiple-choice questions when possible.

Focus areas:

  • What exactly needs to happen?
  • What are the constraints?
  • What's in scope vs out of scope?
  • Are there existing patterns to follow? (consult relevant project skills)
  • What does "done" look like?

2. Explore the codebase

As the discussion progresses, proactively explore relevant code:

  • Search for existing implementations the work relates to
  • Identify files that will likely be affected
  • Note patterns the implementation should follow (reference specific skills)

3. Capture in DISCUSSION.md

Write/update .working/{type}/{slug}/DISCUSSION.md as the conversation progresses:

markdown
# Discussion: {slug}

## Intent
{One paragraph summarizing what the user wants}

## Key Decisions
- **Decision 1**: {What was decided and why}
- **Decision 2**: ...

## Scope
### In scope
- Item 1
- Item 2

### Out of scope
- Item 1

## Relevant Code
- `path/to/file.ts` — {Why it's relevant}
- `path/to/other.ts` — {Pattern to follow}

## Applicable Skills
- `component-architecture` — {Why this skill applies}
- `data-fetching` — {Why this skill applies}
- `styling-design` — {Why this skill applies}
- (only list skills that are relevant to this work)

## Open Questions
- {Any unresolved questions}

## Notes
- {Additional context gathered during discussion}

4. Track assumptions

If any assumptions are made during discussion, create/update .working/{type}/{slug}/ASSUMPTIONS.md:

markdown
# Assumptions: {slug}

- **Assumption**: {What was assumed}
  **Basis**: {Why this seems reasonable}
  **Risk if wrong**: {What would need to change}

5. Wrap up

When the discussion feels complete (scope is clear, decisions are made, no open questions):

  1. Present a summary of DISCUSSION.md to the user
  2. Ask: "Does this capture everything? Ready to move to design with /workflow:design {slug}?"
  3. Update STATUS.md: phase: discussion-complete

Rules

  • One question at a time — never ask multiple questions in one message
  • Multiple choice preferred — easier for the user to respond
  • Capture as you go — don't wait until the end to write DISCUSSION.md
  • Explore proactively — search the codebase to inform questions
  • No implementation — this phase is purely about understanding
  • Track assumptions — if you assume something, write it down