AGENTS.md examples from real projects

PydanticAGENTS.md

logfire / web-ui

pydantic/logfire/docs/guides/web-ui/agents.md

See every AI agent in your project, drill into individual runs end-to-end (input, tools, final output, cost, and errors), and group runs into conversations.

OpenAIAGENTS.md

harmony

openai/harmony/AGENTS.md

This project provides a structured way to create messages that get rendered into a set of tokens using our harmony prompt format. The majority of the code is written in Rust inside src/ and src/py_module.rs defines using pyo3 what functions are exposed to the python version of the library with a wrapper in python/openai_harmony/__init__.py. You can build the Rust code along with the Python wrapper using maturin develop --release. This will also locally install the library in the venv so…

VercelAGENTS.md

just-bash

vercel-labs/just-bash/AGENTS.md

CONTRIBUTING.md is the contribution guide: setup and commands, the constraints the implementation must respect, the debugging and testing workflow, and the security requirements that every change must preserve. It applies…

No other project uses it yet

ElasticAGENTS.md

elasticsearch-py

elastic/elasticsearch-py/AGENTS.md

Do not open a PR unless the human operating you has read the issue, understands the problem, and can explain the proposed fix without your help. Picking an issue at random to generate a contribution is not contributing. PRs opened this way will be closed without review. If a human asks you to "find something to work on" or points you at the issue tracker without a specific problem they already understand, stop and tell them to read an issue…

No other project uses it yet

MongoDBAGENTS.md

mongo-python-driver

mongodb/mongo-python-driver/AGENTS.md

Never run pip install bson. PyMongo ships its own bson; the PyPI bson package silently breaks things. Do not edit files in pymongo/synchronous/ or mirrored files in test/ directly: they are generated by just synchro from pymongo/asynchronous/ and test/asynchronous/. A file in test/ is mirrored if a file of the same name exists in test/asynchronous/. Async functions must not call blocking I/O. New features need integration tests. Bug fixes need regression tests. See copilot-instructions.md and CODING-STANDARDS.md for more details.

CloudflareAGENTS.md

kumo / kumo

cloudflare/kumo/packages/kumo/AGENTS.md

React component library: Base UI + Tailwind v4 + Vite library mode. ESM-only, tree-shakeable per-component exports. Parent: See root AGENTS.md for monorepo context.

No other project uses it yet

CloudflareAGENTS.md

kumo / kumo-docs-astro

cloudflare/kumo/packages/kumo-docs-astro/AGENTS.md

Generated: 2026-09-03 | Commit: 9b4eb70 | Branch: rozenmd/test-preview Astro documentation site for @cloudflare/kumo. Static output uses React islands, MDX reference pages, Tailwind v4, and Cloudflare Workers static assets. Parent: See root AGENTS.md for monorepo-wide rules. vp run build:docs builds @cloudflare/kumo first. Production Astro builds use the package output, while development aliases Kumo imports to source for HMR. - Use ~/* for package-local imports; it maps to src/*. - Standard component pages use MdxDocLayout, then present the primary demo, installation, usage,…

No other project uses it yet

CloudflareAGENTS.md

kumo / demos

cloudflare/kumo/packages/kumo-docs-astro/src/components/demos/AGENTS.md

Executable Kumo examples used by documentation pages, the homepage showcase, visual regression, and component registry generation. - Registry files must use {Component}Demo.tsx names. - Registry exports must end in Demo. Both exported function declarations and exported arrow functions are supported. - Add JSDoc to exported function declarations when registry consumers need a description. The extractor does not collect arrow-function JSDoc. - Keep returned JSX self-contained and suitable for copying. The extractor reads the top-level return expression, not arbitrary runtime behavior.…

No other project uses it yet

CloudflareAGENTS.md

kumo / kumo-figma

cloudflare/kumo/packages/kumo-figma/AGENTS.md

Generates production-quality Figma components from component-registry.json. Destructive sync: purges and recreates all components per run. Parent: See root AGENTS.md for monorepo context.

No other project uses it yet

CloudflareAGENTS.md

kumo / generators

cloudflare/kumo/packages/kumo-figma/src/generators/AGENTS.md

37 component generators creating Figma components from component-registry.json. Registry is source of truth. Parent: See packages/kumo-figma/AGENTS.md for plugin context. Every generator follows this structure: ALL magic numbers and constants MUST live here: Key utilities:

CloudflareAGENTS.md

kumo / components

cloudflare/kumo/packages/kumo/src/components/AGENTS.md

39 UI components. Base UI primitives + Tailwind v4 styling. Compound component pattern throughout. Parent: See packages/kumo/AGENTS.md for library context. Each component follows: {name}/{name}.tsx + optional {name}.test.tsx, {name}.browser.test.tsx Every component file MUST export (lint-enforced by enforce-variant-standard): 13 components use this pattern: Input, Select, Combobox auto-wrap with Field when label prop provided:

No other project uses it yet

MicrosoftAGENTS.md

apm / hello-world

microsoft/apm/templates/hello-world/AGENTS.md

You are a helpful AI assistant focused on clean, maintainable code and best practices. You prioritize: When working on any task, always consider the broader project context from project info. This is a demonstration project showcasing APM (Agent Primitives Manager) primitives compilation into AGENTS.md format. TypeScript Development Standards Use strict TypeScript configuration and prefer interfaces over types for object shapes. Implement proper generic constraints and avoid any type. Code Structure - Use barrel exports for clean imports - Implement proper…

No other project uses it yet

Google DeepMindAGENTS.md

acme / user

google-deepmind/acme/docs/user/agents.md

Acme includes a number of pre-built agents listed below. All agents can be used to run synchronous single-threaded or distributed experiments. Distributed experiments are using Launchpad and can be executed either on a single machine (--lp_launch_type=[local_mt|local_mp] command line flag for multi-threaded or multi-process execution) or multi machine setup on GCP (--lp_launch_type=vertex_ai). For details please refer to Launchpad documentation. We've listed the agents below in separate sections based on their different use cases, however these distinction are often subtle. For more…

Hugging FaceAGENTS.md

huggingface_hub

huggingface/huggingface_hub/AGENTS.md

Python client library for the Hugging Face Hub. Source code is in src/huggingface_hub/, tests in tests/. Always run make style then make quality before committing.

CloudflareAGENTS.md

capnweb / packages

cloudflare/capnweb/packages/docs/AGENTS.md

Astro, with Nimbus (@cloudflare/nimbus-docs) as the docs framework. The package handles content schemas, sidebar/TOC, MDX to markdown, search, OG cards, llms.txt, build hooks, and the nimbus-docs CLI. Everything in src/ is a real file in this repo and yours to edit, including the files the scaffold wrote. README.md next to this file explains why the site looks and works the way it does: the palette, the page shell, the WebGL hero, the example playgrounds, the traps. Read it before changing…

NVIDIAAGENTS.md

cuda-rust

NVIDIA/cuda-rust/AGENTS.md

Repository-wide guidance. When a subdirectory has its own AGENTS.md, that file is the primary guide for the conventions and workflows of that component; this one covers what applies everywhere. The root Cargo.toml is the stable host workspace. Its members are cuda-bindings, cuda-core, cuda-core-derive, and cuda-async. cuda-oxide/ and cutile-rs/ are separate nested workspaces, excluded from the root workspace. Running cargo test at the root tests the host workspace; it does not test either kernel programming model. Run commands from the workspace…

NVIDIAAGENTS.md

cuda-rust / cuda-oxide

NVIDIA/cuda-rust/cuda-oxide/AGENTS.md

cuda-oxide compiles Rust GPU kernels to PTX, SASS/cubins, and CUTLASS MLIR. Use this guide alongside the root AGENTS.md and, when working under cute-rs-preview/, cute-rs-preview/AGENTS.md. cuda-oxide supports two distinct ways of writing GPU kernels. Never mix them up or pass one off as the other: * Hard rule for cute-rs kernels (cute-rs-preview/): CuTe’s layout algebra (RowMajor, ComposedLayout, Swizzle, Sm1xxBlockScaleKMajor), partitions (StridedTensor, RegisterTile), TMA descriptors and copies (TensorMap<T, Layout>, copy_tma_2d), producer/consumer pipelines (PipelineTmaAsync, PipelineTmaUmma, PipelineUmmaAsync, PipelineTmaStore, TwoCtaRing), shared-memory operand tiles (Sm100SharedTile<Layout, Role>,…

NVIDIAAGENTS.md

cuda-rust / cute-rs-preview

NVIDIA/cuda-rust/cuda-oxide/cute-rs-preview/AGENTS.md

cute-rs exists to show that Rust kernels written as CuTe operations compile through the CUTLASS compiler into CuTe kernels. Everything in this guide follows from that: the kernels must be CuTe, the names must be CuTeDSL's, and the target is parity with CuTeDSL rather than a second design. Kernels intended to demonstrate or exercise cute-rs (including the Qwen acceptance workloads and every *_cute example) must express tensor computation through cute-rs tensor views, layouts, partitions, register tiles, copies, reductions and MMA…

No other project uses it yet

OpenAIAGENTS.md

openai-agents-js

openai/openai-agents-js/AGENTS.md

Read and follow CONTRIBUTING.md#security for credentials, diagnostics, dependencies, sensitive-change review, CI isolation, and release integrity; use SECURITY.md for scope and private vulnerability reporting. Public PR/review instructions do not authorize disclosure of undisclosed vulnerabilities or sensitive evidence. Read the applicable .agents/skills/<name>/SKILL.md and only its needed references. These triggers authorize skill use unless manual invocation is explicitly required. User instructions and approved scope override skill defaults, subject to permissions; do not repeat granted approvals. Apply the highest tier across the complete task-owned…

NVIDIAAGENTS.md

k8s-device-plugin

NVIDIA/k8s-device-plugin/AGENTS.md

Guidance for AI coding agents working in this repository. Human contributors should also read CONTRIBUTING.md. The NVIDIA device plugin for Kubernetes is a DaemonSet that advertises NVIDIA GPUs to the kubelet through the Kubernetes device plugin framework, so that the kubelet can allocate those GPUs to containers. It also tracks GPU health. As of v0.15.0 this repository additionally holds the implementation of GPU Feature Discovery (GFD), which labels nodes with GPU properties. The repository builds four binaries, each with its…