How real projects brief Claude Code: the commands, conventions and warnings they put in CLAUDE.md.
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
SentryCLAUDE.mdgetsentry/sentry-rust/CLAUDE.md
Use rust-analyzer-mcp for local code (hover, definitions, diagnostics at file positions). Use crates-mcp for external crate docs. Never use WebFetch for docs.rs or crates.io when these MCPs are available.
MicrosoftReads credentialsCLAUDE.mdmicrosoft/TypeAgent/ts/CLAUDE.md
A fresh checkout of this repo — a new clone, a git worktree (each Copilot CLI session gets its own), or a cloud/container workspace — has the committed files but not the gitignored files that let the repo install and run. Those must be provisioned once. From ts/: On a brand-new machine where pnpm isn't cached yet, pnpm run setup can't bootstrap itself (corepack would try to fetch pnpm from the blocked public registry). Start with the pnpm-free entrypoint instead…
MicrosoftCLAUDE.mdmicrosoft/TypeAgent/ts/examples/planValidationDemo/CLAUDE.md
This project is governed by a plan-validation MCP server. Every file access and shell call you make must go through the validated_* tools. The built-in Read, Write, Edit, Bash, Glob, Grep tools are NOT available in this session — only the MCP tools provided by the plan-validation server. A live dashboard is running at http://localhost:3100 — every tool call you make is visible there, including blocks. Call get_plan_schema. Read it carefully — the AgentPlan DSL has strict shape requirements and…
MicrosoftCLAUDE.mdmicrosoft/TypeAgent/ts/packages/agents/powershell/CLAUDE.md
PowerShell captures and reuses PowerShell scripts from reasoning traces. Scripts persist in instance storage (~/.typeagent/profiles/<profile>/powershell/) across sessions. Grammar rules are registered dynamically at runtime — no build step needed for user-created flows. Initialize flow storage and seed samples only when the root powershell schema is enabled. Static namespaces can enable concurrently; they must not create competing store instances or revoke the root flow approvals on disable. Dynamic flows require trusted user authorization and the Windows execution broker, without execution feature…
MicrosoftCLAUDE.mdmicrosoft/TypeAgent/ts/packages/agents/taskflow/CLAUDE.md
TaskFlow runs user-taught multi-step macros over existing TypeAgent actions. Flows are pure JSON — no TypeScript logic. A deterministic compile script handles all boilerplate generation (TypeScript type, grammar rule, manifest entry). When asked to "compile pending recipes" or similar: That's it. scripts/compileRecipes.mjs reads pending/*.recipe.json and: If the build fails, fix the error in the generated files and re-run pnpm run build. When asked to write a recipe for a new flow (or if you notice a missing flow): Do NOT…
AWSCLAUDE.mdawslabs/automated-security-helper/ash-agent-plugins/agentic-coding/plugins/claude/CLAUDE.md
@../AGENTS.md
AnthropicCLAUDE.mdanthropics/cwc-long-running-agents/claude-code-config/.claude/CLAUDE.md
The verify-gate hook will deny writes to test-results.json until you have opened evidence. Do not try to work around it.
VercelReads credentialsCLAUDE.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. Authentication (controller/auth.ts) is built from the value passed to createClient: The Controller's unauthorized flag gates usage tracking and skips the final blocking fetch during initialization. It is set on an auth failure from any source (stream init, poll, one-time fetch), meaning a 401 or a missing OIDC token (OidcTokenError from auth.ts). @vercel/oidc also refreshes tokens over the network and flattens every failure into…
MetaReads credentialsCLAUDE.mdfacebookresearch/MLGym/CLAUDE.md
MLGym is a framework for benchmarking AI research agents on machine learning tasks. Key components:
NVIDIACLAUDE.mdNVIDIA/TensorRT-Edge-LLM/CLAUDE.md
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. -->
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.
Google DeepMindCLAUDE.mdgoogle-deepmind/simply/CLAUDE.md
Simply is a minimal JAX-based research codebase for LLM training and inference. It emphasizes minimal abstractions for rapid iteration on frontier research. The codebase supports Gemma, Qwen, and DeepSeek model families with multi-host distributed training.
VercelReads credentialsCLAUDE.mdvercel/workflow-examples/custom-adapter/CLAUDE.md
Use Bun instead of Node.js, npm, pnpm, or vite.
MicrosoftCLAUDE.mdmicrosoft/WindowsProtocolTestSuites/CLAUDE.md
This is the Windows Protocol Test Suites repository — a .NET-based interoperability testing framework for Microsoft Open Specification protocols. Follow these steps before starting any task: Use kebab-case for the…
MetaCLAUDE.mdfacebookresearch/spider/CLAUDE.md
SPIDER (Scalable Physics-Informed DExterous Retargeting) is a physics-based retargeting framework that converts human motion (from video or mocap) into robot actions for dexterous hands and humanoid robots. Developed by Meta/FAIR.…
AnthropicCLAUDE.mdanthropics/connect-rust/CLAUDE.md
@CONTRIBUTING.md Before producing a commit, run both review agents in parallel (single message, two Agent tool calls): - rust-code-reviewer — correctness, safety, ownership/lifetimes, performance - rust-api-ergonomics-reviewer — downstream-consumer perspective: happy-path friction, lints that fire in user crates, runtime footguns, doc drift, generated-code readability The two are complementary (different lenses on the same diff) and produce largely non-overlapping findings. Address all Critical, High, and Medium findings from both. For Low / advisory findings, flag these to the user to decide. For…
ElasticCLAUDE.mdelastic/elasticsearch-java/CLAUDE.md
See AGENTS.md for guidance on working with code in this repository.
HashiCorpCLAUDE.mdhashicorp/design-system/CLAUDE.MD
Read AGENTS.md
GoogleCLAUDE.mdgoogle/xrblocks/CLAUDE.md
See GEMINI.md.
NVIDIACLAUDE.mdNVIDIA/tilus/CLAUDE.md
Tilus caches generated kernels. During development, set the cache directory via tilus.option.cache_dir("some-cache-dir") to inspect the generated source.cu files. drivers.py:build_program orchestrates the full compilation: 1. Verify — ir.tools.verify(prog) 2. Optimize (Tilus IR) — optimize_program applies get_default_passes() from transforms/__init__.py 3. Lower to Hidet IR — backends.codegen.generate_ir_module 4. Optimize (Hidet IR) — optimize_ir_module applies Hidet-level passes 5. Codegen — Emit CUDA C source 6. Compile — Build .so via nvcc
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.
