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。
部署与云服务

superclaude

Configure and operate the Claude Code harness for large codebases. Builds CLAUDE.md hierarchies, scoped test/lint commands, file exclusions, codebase maps, hooks, skills, subagent strategies, and LSP/MCP wiring. Use when setting up Claude Code for a new repo, auditing an existing configuration, onboarding a team, or scaling from single-developer to org-wide deployment. Triggers on "set up Claude Code for this repo", "optimize my Claude Code config", "audit my CLAUDE.md", "make this codebase navigable", "configure hooks/skills/plugins".

av · 已有静态检查记录

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 av/skills --skill superclaude

技能说明


name: superclaude description: Configure and operate the Claude Code harness for large codebases. Builds CLAUDE.md hierarchies, scoped test/lint commands, file exclusions, codebase maps, hooks, skills, subagent strategies, and LSP/MCP wiring. Use when setting up Claude Code for a new repo, auditing an existing configuration, onboarding a team, or scaling from single-developer to org-wide deployment. Triggers on "set up Claude Code for this repo", "optimize my Claude Code config", "audit my CLAUDE.md", "make this codebase navigable", "configure hooks/skills/plugins".

Superclaude

Configure the Claude Code harness so it works as well as the model. The harness — CLAUDE.md files, hooks, skills, plugins, LSP, MCP servers, subagents — determines whether Claude Code is productive or fumbling. This skill builds and tunes that harness for large codebases.

When to Load

  • Setting up Claude Code for a new repo or monorepo
  • Auditing an existing Claude Code configuration
  • Onboarding a team or org onto Claude Code
  • Codebase navigation is slow or inaccurate
  • Claude keeps opening wrong files, running wrong tests, or missing context
  • Scaling from one developer to many

The Harness

Seven components, loaded in priority order. Each has a specific job. Misplacing work between them is the most common configuration mistake.

#ComponentLoadsBest ForCommon Mistake
1CLAUDE.mdEvery sessionProject conventions, codebase layout, critical gotchasStuffing reusable expertise that belongs in a skill
2HooksOn eventsDeterministic automation, capturing learnings, enforcing rulesUsing prompts for things that should run automatically
3SkillsOn demandReusable expertise, specialized workflowsLoading everything into CLAUDE.md instead
4PluginsAlways, once configuredDistributing org-wide setups to all developersLetting good setups stay tribal knowledge
5LSPAlways, once configuredSymbol-level navigation, typed languagesAssuming it works automatically without setup
6MCP serversAlways, once configuredInternal tools, data sources, APIsBuilding MCP connections before basics work
7SubagentsWhen invokedParallel exploration, splitting read from writeRunning exploration and editing in the same session

The rule: Start from the top. Get CLAUDE.md right before touching hooks. Get hooks right before building skills. Don't wire up MCP servers when your CLAUDE.md still says run npm test for a monorepo with 40 services.


1. CLAUDE.md — The Foundation

CLAUDE.md is loaded every session. It is the most impactful configuration surface. It is also the easiest to get wrong.

What Goes In

  • Pointers and critical gotchas only
  • Codebase structure overview (what lives where)
  • Build/test/lint commands scoped to directories
  • Conventions that differ from language defaults
  • Hard constraints (never modify X, always use Y for Z)

What Stays Out

  • Reusable expertise (belongs in a skill)
  • Long tutorials or documentation dumps
  • Things the model already knows (standard language idioms, common framework patterns)
  • Anything that hasn't changed in months and isn't surprising

Hierarchical Strategy

CLAUDE.md files are additive. Claude walks up the directory tree and loads every CLAUDE.md it finds. Use this.

repo/
  CLAUDE.md                  # Repo-wide: architecture, top-level pointers, critical rules
  services/
    CLAUDE.md                # Services area: shared patterns, common dependencies
    payments/
      CLAUDE.md              # Payments service: specific test commands, API conventions
    auth/
      CLAUDE.md              # Auth service: security constraints, token handling rules
  frontend/
    CLAUDE.md                # Frontend: build tool, component patterns, styling approach

Root-level CLAUDE.md gets the big picture: directory structure overview, overarching conventions, cross-cutting gotchas. Keep it under 100 lines. Everything else loads on demand.

Subdirectory CLAUDE.md gets local specifics: test commands, lint commands, local conventions, dependencies, gotchas for that area. A developer working in services/payments/ gets the root context automatically plus the payments-specific context.

Scoped Test and Lint Commands

Running the full test suite when Claude changed one file in one service causes timeouts and wastes context. Scope commands per directory.

Root CLAUDE.md:

# Testing
Do not run the full test suite. Each service has its own test command in its CLAUDE.md.

services/payments/CLAUDE.md:

# Testing
Run: `cd services/payments && npm test`
Single file: `cd services/payments && npm test -- --testPathPattern=<file>`
Lint: `cd services/payments && npm run lint`

For compiled-language monorepos with cross-directory dependencies, document the build graph in the root CLAUDE.md so Claude knows which targets to rebuild.

Codebase Maps

When the repo has dozens or hundreds of top-level directories, add a map. Not a novel — a table of contents.

Root CLAUDE.md:

# Directory Structure
- `services/` — Backend microservices (payments, auth, notifications, inventory)
- `frontend/` — React SPA, shared component library
- `infra/` — Terraform modules, CI/CD pipelines
- `libs/` — Shared libraries (logging, metrics, auth-client)
- `scripts/` — Build and deploy tooling
- `proto/` — Protobuf definitions, generated code

For massive repos, use layered maps: root describes the top level, each area's CLAUDE.md describes its subdirectories. Claude loads the next level of detail only when it enters that directory.

Writing the Root CLAUDE.md

When setting up a new repo, follow this template:

# <Project Name>

## Architecture
<2-3 sentence description of what this project is and how it's structured>

## Directory Structure
<ma
阅读完整来源 ↗