菜单
Design a feature before writing code. Spawns planner + skeptic subagents that read the Memory Bank, argue, and write a plan to swarm-report/. Use BEFORE /build, for any feature touching 2+ files or with unclear scope. Skip for typos and one-line fixes.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ayusavin/labs-harness --skill plan命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Turn a design (Figma file or an exported reference deck) into a build ladder — inventory every frame, extract tokens and atoms, find the gaps the designer still owes you, and write a /build-ready plan plus the .claude/design-gate.json rungs that /visual-verify will check. Use BEFORE writing any UI code from a design. Not for features without a design.
Install this mini dev-loop harness into a target project. Copies the agents, skills (plan/build/review/debug), the test-gate hook, settings, and AGENTS.md into the given project path, then seeds a Memory Bank index if none exists. Run this from a clone of the harness repo — pass the absolute path to the project you want to equip.
Check the built UI against the design, rung by rung — capture the real rendered component (Storybook via Playwright on web, the debug preview gallery on an iOS simulator or Android emulator), diff it against the 1:1 design export, and write the verdict the visual-gate Stop hook reads. Use after /build lands a rung, and whenever the gate blocks. Requires .claude/design-gate.json.
基于 SOC 职业分类
正在显示 SKILL.md
| name | plan |
| description | Design a feature before writing code. Spawns planner + skeptic subagents that read the Memory Bank, argue, and write a plan to swarm-report/. Use BEFORE /build, for any feature touching 2+ files or with unclear scope. Skip for typos and one-line fixes. |
You are the ORCHESTRATOR. You do not design the plan yourself — you run two subagents and merge their output. No code edits here.
/plan "<one-sentence feature>"
Slugify the feature → <slug> (kebab-case, short).
Read context yourself (cheap, keeps subagents focused): .memory-bank/index.md
if it exists. If it does not, tell the user: "No Memory Bank found — plan will rest
only on your description." and continue.
Spawn 2 subagents in parallel — one message, two Task calls,
subagent_type: general-purpose. Prefix EACH prompt with:
Answer TERSE: terse, drop filler, keep all technical substance.
.claude/agents/planner.md and follow it exactly. Feature: .".claude/agents/skeptic.md and follow it exactly. Feature: ."(Sharper, slightly pricier variant: run planner first, then pass its YAML into the skeptic prompt so it critiques the concrete plan. One extra round-trip. Use it when the feature is risky.)
Merge (orchestrator only): take the planner's plan, apply the skeptic's HIGH/MED fixes, and list anything unresolved under Blockers.
Write swarm-report/<slug>-plan.md:
# Plan: <feature> (slug: <slug>)
## TL;DR
## Acceptance criteria — observable "done" conditions; /review checks these
## Plan — merged steps + affected files + tests
## Blockers — skeptic HIGH not yet resolved (human decides)
## Out of scope
## Assumptions
Report to user: TL;DR + any Blockers. If Blockers exist, ask the user to resolve
them before /build.