AgentSkillsCN

team-develop

当您启动一项新功能、进行重新设计,或实施重大变更且此类调整有望受益于多角色协作时,此技能便能大显身手。您可以组建一支虚拟 IT 团队——产品经理、分析师、用户体验设计师、质量保证工程师、软件工程师——并通过用户验收点,将各角色有机串联,形成高效的协作流水线。

SKILL.md
--- frontmatter
name: team-develop
description: "Use when starting a new feature, redesign, or significant change that benefits from multi-role collaboration. Assembles a virtual IT team (Product Engineer, Analyst, UX Designer, QA Engineer, Software Engineer) and orchestrates them as a pipeline with user checkpoints."

Team Develop — Virtual IT Team Orchestration

Overview

Assemble a virtual IT team to take a feature from idea to implementation. You are the Team Lead — you orchestrate, delegate, and ensure quality. You NEVER write code yourself.

The team works as a pipeline: each role produces artifacts that feed into the next. The user reviews and approves after every stage.

When to Use

  • New feature implementation requiring multiple perspectives
  • Significant UI/UX redesign
  • Complex refactoring with product implications
  • Any task that benefits from product thinking + decomposition + UX + QA + implementation

Team Roles

RoleFocusMandatory Skill
Product EngineerProduct value, user stories, acceptance criteriasuperpowers:brainstorming
AnalystTask decomposition, file paths, dependenciessuperpowers:writing-plans
UX DesignerUser experience, interaction flows, accessibilityweb-design-guidelines
QA EngineerDefinition of Done, test scenariossuperpowers:verification-before-completion
Software EngineerImplementation, TDD, commitssuperpowers:test-driven-development

The Pipeline

dot
digraph pipeline {
    rankdir=TB;
    node [shape=box];

    discovery [label="Phase 0: Discovery\n(brainstorming skill)"];
    product [label="Phase 1: Product Engineer"];
    analyst [label="Phase 2: Analyst"];
    ux [label="Phase 3: UX Designer"];
    qa [label="Phase 4: QA Engineer"];
    engineer [label="Phase 5: Software Engineer"];
    validation [label="Phase 6: QA + UX Validation"];

    cp0 [label="CHECKPOINT" shape=diamond];
    cp1 [label="CHECKPOINT" shape=diamond];
    cp2 [label="CHECKPOINT" shape=diamond];
    cp3 [label="CHECKPOINT" shape=diamond];
    cp4 [label="CHECKPOINT" shape=diamond];
    cp5 [label="CHECKPOINT" shape=diamond];
    cp6 [label="CHECKPOINT" shape=diamond];

    discovery -> cp0 -> product -> cp1 -> analyst -> cp2 -> ux -> cp3 -> qa -> cp4 -> engineer -> cp5 -> validation -> cp6;

    cp6 -> engineer [label="issues found" style=dashed];
}

Phase 0: Discovery (REQUIRED)

You MUST invoke the superpowers:brainstorming skill and follow it exactly.

Process:

  1. Explore the project context (files, docs, tech stack, recent commits)
  2. Follow brainstorming process: ask questions one at a time, multiple-choice preferred
  3. Propose 2-3 approaches with trade-offs
  4. Present design in sections (200-300 words), validate each with user
  5. Save validated design to docs/plans/YYYY-MM-DD-<feature>-design.md
  6. Extract brief:
    • Goal: one sentence
    • Problem statement: what user problem this solves
    • Affected area: which parts of codebase
    • Constraints: technical, timeline, backward compatibility
    • Success criteria: how we know it's done

CHECKPOINT: Present the brief to user. Wait for approval before continuing.

Phases 1-4: Specialist Agents

For each phase, follow this exact sequence:

Step 1: Read the role prompt

Read the corresponding role-*.md file from this skill's directory.

Step 2: Dispatch the agent

Use Task tool (subagent_type: general-purpose) with:

  • The role prompt template
  • The brief from Phase 0
  • ALL artifacts from previous phases (full text, not file references)
  • Instruction to discover and apply relevant skills (see Skill Discovery below)

Step 3: Receive and save output

  • Save the agent's output to docs/plans/YYYY-MM-DD-<feature>-<role>.md
  • Present a summary to the user

Step 4: CHECKPOINT

Show the summary. Ask user: "Approve, request changes, or skip this stage?"

  • Approve: continue to next phase
  • Changes: describe what to change, re-dispatch the same role
  • Skip: mark as skipped, continue with available context

Phase 5: Implementation

Read role-software-engineer.md. For EACH task from the Analyst's plan:

  1. Dispatch a fresh Software Engineer subagent with:
    • The specific task description
    • Product spec + UX criteria + test plan for that task
    • Instruction to follow TDD
  2. Receive implementation report
  3. CHECKPOINT after each task: show files changed, tests written, commits made

Phase 6: Validation

  1. Dispatch QA validator (from role-validators.md) with:
    • Original test plan
    • Implementation report from all tasks
    • Instruction: verify DoD compliance independently, do NOT trust the engineer's report
  2. Dispatch UX validator (from role-validators.md) with:
    • Original UX criteria
    • Implementation report
    • Instruction: verify UX compliance from user's perspective

If issues found:

  • Route specific issues back to Software Engineer for fixes
  • Re-validate after fixes
  • Max 3 reroute cycles, then escalate to user

CHECKPOINT: Present final validation report.

Skill Discovery (REQUIRED for every agent)

Every agent prompt MUST include this block:

code
Before you begin your work:
1. Search for available skills:
   - Look in ~/.claude/skills/ for SKILL.md files
   - Look in ~/.claude/plugins/**/skills/ for SKILL.md files
2. Read the name and description of each found skill
3. Apply any skill that is relevant to YOUR role
4. List which skills you applied in your report

Context Accumulation Rule

Each phase receives ALL artifacts from prior phases. Never summarize — pass full text.

code
Phase 1 receives: brief + design
Phase 2 receives: brief + design + product spec
Phase 3 receives: brief + design + product spec + task plan
Phase 4 receives: brief + design + product spec + task plan + UX criteria
Phase 5 receives: per-task slice of all above
Phase 6 receives: all above + implementation reports

Artifacts Produced

All saved to docs/plans/ in the project directory:

PhaseFile
0YYYY-MM-DD-<feature>-design.md
1YYYY-MM-DD-<feature>-product-spec.md
2YYYY-MM-DD-<feature>-tasks.md
3YYYY-MM-DD-<feature>-ux-criteria.md
4YYYY-MM-DD-<feature>-test-plan.md
6YYYY-MM-DD-<feature>-validation.md

Edge Cases

SituationAction
User rejects checkpointAsk what to change. Re-dispatch same role with corrections.
User wants to skip a roleAllow it. Mark as skipped, continue with available context.
Agent asks questionsRelay to user, collect answer, resume agent.
Validation fails 3 timesStop rerouting. Present issues to user, ask how to proceed.
No docs/plans/ directoryCreate it automatically.
Codebase has no testsSoftware Engineer creates test infrastructure as Task 0.

Red Flags — STOP and Escalate to User

  • You are about to write implementation code yourself
  • An agent produced output that contradicts the user's approved design
  • Two agents produced conflicting requirements
  • Validation cycle exceeded 3 iterations
  • Agent reports it cannot complete the task

What This Skill Does NOT Do

  • Does not push to git remote (asks user first)
  • Does not create PRs (defers to superpowers:finishing-a-development-branch)
  • Does not make architectural decisions without user approval
  • Team Lead NEVER writes code — strictly delegates

Integration

Required skills:

  • superpowers:brainstorming — Phase 0 discovery
  • superpowers:writing-plans — Phase 2 task decomposition

Recommended skills:

  • superpowers:test-driven-development — Phase 5 implementation
  • superpowers:verification-before-completion — Phase 4+6 QA
  • superpowers:systematic-debugging — Phase 5 if issues arise
  • superpowers:using-git-worktrees — isolated workspace setup
  • superpowers:finishing-a-development-branch — after Phase 6
  • web-design-guidelines — Phase 3 UX validation