How real projects brief Claude Code: the commands, conventions and warnings they put in CLAUDE.md.
tuannvmCLAUDE.mdtuannvm/codex-mcp-server/CLAUDE.md
This is an MCP (Model Context Protocol) server that wraps OpenAI's Codex CLI, exposing it as tools to Claude Code and other MCP clients. Sessions enable multi-turn conversations with Codex:
Matt-DionisCLAUDE.mdMatt-Dionis/claude-code-configs/claude-config-composer/CLAUDE.md
This configuration combines: Next.js 15, shadcn/ui Combined from: Next.js 15, shadcn/ui This is a Next.js 15 application using: - App Router (not Pages Router) - React 19 with Server Components by default - TypeScript for type safety - Tailwind CSS for styling (if configured) - Server Actions for mutations - Turbopack for faster builds (optional) This is a shadcn/ui project focused on: - Component-first development with copy-paste architecture - Radix UI primitives for behavior and accessibility - Tailwind CSS for…
VercelCLAUDE.mdvercel/flags/packages/vercel-flags-core/CLAUDE.md
Core feature flag evaluation engine for Vercel. Handles flag evaluation, real-time updates via streaming, and usage tracking. Main interface for interacting with flags:
chunxiaoxxCLAUDE.mdchunxiaoxx/nautilus-compass/CLAUDE.md
🔴 开工第一动作(frame-upgrade-20260904-compass #2 · MetaGPT _observe 原则):先查平台信箱再开始其他任何事——curl -s "https://nautilus.social/api/platform/org/mailbox?to=compass&unread=1" → 逐封 ack(需正文回函的按 trace 回函),然后才开工。本框状态自报见 FRAME.yaml(状态变化时顺手改)。 🔴 方向锚(2026-09-02 用户拍,与仓内任何文档冲突时以此为准): 当前唯一业务 = 具身智能数据采集 + 数据飞轮 + 第三方数据有效性验证平台; 垂域出题/交付业务已死不再做——仓内 CHARTER/SSOT 等历史层的垂域/FDE 燃料内容全是 8…
teng-linCLAUDE.mdteng-lin/weread-omni/CLAUDE.md
No summary in the file. Open it to read it.
VercelCLAUDE.mdvercel-labs/bash-tool/CLAUDE.md
This file provides guidance for Claude Code when working on this project. bash-tool is a generic bash tool for AI agents, compatible with AI SDK (https://ai-sdk.dev/). It provides bash, readFile, and writeFile tools that work with a sandboxed environment. Tests are co-located with source files (*.test.ts). Tests are included in TypeScript compilation for type checking. AI SDK tool execute calls require ToolExecutionOptions with toolCallId and messages.
emiliaprotocolCLAUDE.mdemiliaprotocol/emilia-protocol/CLAUDE.md
No summary in the file. Open it to read it.
sno-aiCLAUDE.mdsno-ai/mda/CLAUDE.md
Project memory for Claude Code. Read on every session. No more asking the user about version numbers, npm tokens, trusted publishing, or auth. Everything below is settled. Just execute. 1. Edit apps/cli/package.json → bump version 2. pnpm cli:test (must pass; tests dynamically read the version) 3. git add apps/cli/package.json && git commit -m "chore(cli): bump to <version>" 4. git push origin main 5. git tag v<version> HEAD 6. git push origin v<version> 7. gh run watch $(gh run list --workflow=publish-cli.yml…
dislerCLAUDE.mddisler/infinite-agentic-loop/CLAUDE.md
This is an experimental project demonstrating the Infinite Agentic Loop pattern using Claude Code's custom slash commands. The project orchestrates multiple AI agents in parallel to generate evolving iterations of…
ggozadCLAUDE.mdggozad/haiku.rag/CLAUDE.md
The project guide is AGENTS.md. This file carries only what applies to Claude Code. Two skills live in .claude/skills/ and drive the Logfire MCP: - debug-evals — eval runs (servicename='evals'): recent runs, per-case pass rate / citedmap, failing and slowest cases. - debug-ingestion — ingestion (service_name='haiku-ingester'): failed/dead jobs, docling-serve failover, per-source sweeps, worker breaker trips. Both start from mcplogfirequeryschemareference and query the records table via mcplogfirequery_run. Ingester stdlib logs (dead/reschedule narration) are NOT shipped to Logfire — only spans and…
carlos-algmsCLAUDE.mdcarlos-algms/agentic.nvim/CLAUDE.md
CRITICAL See @AGENTS.md for Instructions, it's the source of truth for all my instructions, no exceptions, it should be read and be part of your context before any other action.
ee0pdtCLAUDE.mdee0pdt/Godot-MCP/CLAUDE.md
No summary in the file. Open it to read it.
HELPMEEADICECLAUDE.mdHELPMEEADICE/doge-code/CLAUDE.md
There is currently no first-class root lint script or automated test script in package.json. There is also no documented single-test command at the repository root, so validation is primarily targeted…
dbt-labsCLAUDE.mddbt-labs/dbt-mcp/CLAUDE.md
dbt-mcp is an MCP (Model Context Protocol) server that exposes dbt functionality as tools to AI assistants. Built on FastMCP from the mcp SDK. Tools follow a consistent pattern: 1. @dbtmcptool decorator defines the tool with metadata 2. ToolName enum in tools/toolnames.py — every tool needs an entry 3. Toolset mapping in tools/toolsets.py — maps tools to categories 4. Context injection via adaptcontext() — tools receive typed context objects, but MCP only sees user-facing params 5. register_tools() in tools/register.py —…
majiayu000CLAUDE.mdmajiayu000/claude-skill-registry/CLAUDE.md
This project is a three-repo split. Before editing anything, identify which repo you are in and route the change to the owner repo instead of patching generated outputs in the wrong place.
majiayu000CLAUDE.mdmajiayu000/claude-skill-registry/skills/CLAUDE.md
This repository is the archive leg of a three-repo split: - claude-skill-registry-core: source of truth for workflows, scripts, indexing, Pages, and publish orchestration - claude-skill-registry-data: archived skill contents - claude-skill-registry: merged publish artifact
rome-osCLAUDE.mdrome-os/rome/docs/CLAUDE.md
Durable design documentation for Rome. README.md maps what lives where, and the per-content-type rulebooks live in authoring/. A section that names a mechanism instead of an invariant reads as documentation and cannot be applied to a diff. Each section in concepts/ and architecture/ names an invariant between components — a rule a contributor uses to decide whether a change is allowed — statable without reference to any file, type, or function. The routing and agent layers cannot observe which external…
rome-osCLAUDE.mdrome-os/rome/rome_apps/CLAUDE.md
Each romeapps/<appId>/ is a runtime-loaded app (plugin), shipped first-party with Rome. They use the same app model as user-authored apps. The canonical authoring docs below are shared with the in-Rome appcreation skill, so the host coding agent and the in-Rome agent follow the same instructions. An app that reads a host CSS variable it never compiled renders correctly here and nowhere else. The host mounts each app's web bundle in a Shadow DOM, and custom properties inherit across that boundary.…
adalineCLAUDE.mdadaline/gateway/CLAUDE.md
Adaline Gateway is a production-grade unified SDK ("Super SDK") for calling 300+ LLMs through a single TypeScript interface. It is fully local (not a proxy), strongly typed, and isomorphic. It supports batching, retries, caching, callbacks, OpenTelemetry, and tool calling. Repository: https://github.com/adaline/gateway Per-package (run from package directory): When adding a new provider, follow existing providers (e.g., core/providers/openai/) as the template. Internal dependencies use workspace:* protocol. Commits follow conventional commits enforced by commitlint: - Format: type(scope): description - Types: feat, fix, chore,…
FerroxLabsCLAUDE.mdFerroxLabs/wayland/CLAUDE.md
No summary in the file. Open it to read it.
A file Claude Code reads at the start of every session. It holds the commands, conventions and warnings the agent needs for this project.
At the repository root. Claude Code also reads CLAUDE.md files in subdirectories when it works there.
Build and test commands, the project's layout, conventions that aren't obvious from the code, and mistakes to avoid. Short files tend to work better than long ones.
Claude Code reads CLAUDE.md; most other agents read AGENTS.md. Many projects keep one and point the other at it.
