Microsoft Agent 365 DevTools CLI (a365) - A .NET CLI tool built on .NET 8.0 with support for running on .NET 8.0 or higher (e.g., .NET 9, 10). Used for deploying and managing Microsoft Agent 365 applications on Azure. Supports .NET, Node.js, and Python applications with auto-detection. Test Framework: xUnit with FluentAssertions and NSubstitute Test Location: src/Tests/Microsoft.Agents.A365.DevTools.Cli.Tests/ Parallel Test Execution: Tests modifying environment variables or shared resources must disable parallelization:
dynwinrt is a Rust-based runtime library that enables dynamic invocation of Windows Runtime (WinRT) APIs. Unlike static projections (PyWinRT, C++/WinRT), this library uses runtime metadata (.winmd files) and FFI (libffi) to call arbitrary WinRT methods without native code generation. It provides JavaScript (napi-rs) and Python (PyO3) bindings, plus a code generation tool (dynwinrt-codegen) that produces typed wrappers from .winmd files. Classic COM and flat Win32 have separate JavaScript/TypeScript projections and runtime subpaths. Flat Win32 DLL exports use an independent native…
Pure-Python SMB3 server for quicksand host-guest directory mounting. An SMB3 file server with two transports. On macOS and Linux it runs inetd-style, reading/writing on stdin/stdout — QEMU's guestfwd=cmd: spawns one instance per guest TCP connection, and no TCP port is opened on the host, eliminating the attack surface that Samba's smbd had. On Windows, quicksand-core serves it in-process via serve_socket on a persistent loopback-only (127.0.0.1) TCP listener, which avoids requiring Administrator rights. The Linux kernel CIFS client always performs a…
monosize is a CLI tool that measures bundle sizes for packages and monorepos and compares them against a stored baseline (typically main). This repo is itself a Yarn 4 + Nx monorepo of the CLI plus first-party adapter packages. The packages/ directory has three roles: Adapters are wired up by users in their own monosize.config.mjs; this repo does not consume itself end-to-end. When changing the BundlerAdapter / StorageAdapter shape, every adapter package must be updated in lockstep. Per-package and full-graph…
Microsoft Agent 365 SDK for Python - Enterprise-grade extensions for building production-ready AI agents for M365, Teams, Copilot Studio, and Webchat. This is a multi-package monorepo organized as a uv workspace with 13 interdependent packages. Python Version: 3.11+ (3.11 and 3.12 tested in CI) For detailed architecture and design documentation, see docs/design.md, which includes: - Complete package descriptions and usage examples - Design patterns (Singleton, Context Manager, Builder, Result, Strategy) - Data flow diagrams for agent invocation tracing and MCP…
Never hardcode model architecture names, node/operator names, input/output tensor names, layer naming patterns, or any model-specific logic. All solutions must be universal and architecture-agnostic. All testing uses pytest with code-generated results. Never create standalone test scripts, use LLM-generated expectations, or generate test data manually. Run the necessary pytest scope after every implementation or test revision (for example, affected test files/modules/cases). Run full suite uv run pytest tests/ only when explicitly requested, before merge/release, or when impact is broad/unclear. Never assume…
Inherits all rules from /CLAUDE.md. Additional test-specific rules below. Reference: /docs/pytest-best-practices.md - Place unit tests under tests/unit/<module>/ mirroring src/winml/modelkit/<module>/ - Place integration tests under tests/integration/, e2e under tests/e2e/ - Place cross-cutting CLI-surface tests (startup, import budget, arg parsing, command discovery, version/help output) under tests/cli/ — these don't mirror any single src/ module, so they don't fit under tests/unit/<module>/ - Put shared fixtures in the narrowest conftest.py that covers all consumers - Place tests that mirror a src/winml/modelkit/<module>/ directly under tests/…
Important: This file (and all sub-project CLAUDE.md files) are checked into the repository. Only include information that is valid for any developer or machine: project conventions, architecture, commands, constraints. Do not add machine-specific paths, personal tool preferences, local environment settings, or anything that would not apply to every contributor. The CI runs on windows-latest and builds for multiple targets: wasm32-wasip1, aarch64-unknown-linux-musl, x8664-pc-windows-msvc, x8664-unknown-linux-gnu. The release build uses Spectre mitigations (/Qspectre /sdl) and produces preflaters07.dll and preflateutil.exe. preflate-rs analyzes DEFLATE-compressed streams,…
Scans binary files (ZIP, PNG, JPEG) for DEFLATE streams, orchestrates the preflate + Zstd pipeline, and reassembles the output. Only format version 2 exists (v1 was removed). PreflateContainerConfig holds knobs: minchunksize, maxchunksize, totalplaintextlimit, chunkplaintextlimit, validatecompression, maxchain_length. All framing bytes (type, varint) are written directly to the output stream — they are never inside the Zstd encoder. Each block type byte encodes two fields: Mask constants (defined in container_processor.rs): - A single persistent zstd::stream::write::Encoder is created once and shared across all…
C-compatible DLL for .NET interop. Exposes a streaming compress/decompress API as extern "C" functions. The version number is baked into the crate name (preflaters0_7) for binary compatibility. flags encoding: - bits 0–4: Zstd compression level - bit 5: test_baseline - bit 6: verify Return value of compress_buffer: 0 = more output available, 1 = done, <0 = error. Magic numbers are validated on every call to catch dangling/wrong pointer bugs. - Uses #[unsafe(nomangle)] on exported functions — the only place…
libfuzzer harnesses for fuzzing the core and container APIs. Not published; requires the fuzzing cargo feature. Feeds arbitrary bytes to preflatewholedeflatestream() as a raw DEFLATE stream, then attempts recreatewholedeflatestream() on the result. Verifies no crash or panic. Feeds arbitrary bytes (minimum 1 byte) to preflatewholeintocontainer(), then recreatewholefromcontainer(), and asserts the output matches the original input.
Core DEFLATE analysis and reconstruction. Analyzes a DEFLATE bitstream, extracts the uncompressed plaintext plus a compact set of reconstruction parameters, and later recreates the bit-exact original bitstream. PreflateConfig controls maxchainlength, plaintextlimit, and verify_compression. Reconstruction runs the same pipeline in reverse.
End-to-end round-trip tests against real sample files in samples/. - Core round-trip: decompress a .deflate file with preflatewholedeflatestream, recompress with recreatewholedeflatestream, assert bitwise identical output. - Container round-trip: compress a ZIP / PNG / DOCX / PDF through PreflateContainerProcessor, decompress with RecreateContainerProcessor, assert the output matches the original file byte-for-byte. The samples/ directory contains real-world compressed files used as test fixtures: deflate streams, zlib streams, PNGs, ZIPs, PDFs, DOCX, JPEG, WebP, and binary blobs from various compressors (zlib, zlib-ng, libdeflate,…
CLI tool for manually testing preflate compression on files and directories. <PATH> may be a single file or a directory (scanned recursively). Single file: src/main.rs (~193 lines). Helper asserteqarray<T>() provides detailed positional diff output for debugging mismatches during verification.
Tech Stack: Electron 35 + React 18 + TypeScript 5, Webpack 5, TailwindCSS 4, lucide-react icons, Vercel AI SDK 5.x (streaming), @modelcontextprotocol/sdk, better-sqlite3 (local persistence), Monaco Editor, Playwright (browser automation). Architecture: Electron multi-process model — main process (Node.js: auth, chat engine, MCP runtime, data persistence, voice) + renderer process (React SPA: chat UI, agent editor, settings) + preload scripts (type-safe IPC bridge). The app is built as OpenKosmos-only. A list of useful package.json scripts: After entering a module directory, check…
This section outlines the core implementation philosophy and guidelines for software development projects. It serves as a central reference for decision-making and development approach throughout the project. Embodies a Zen-like minimalism that values simplicity and clarity above all. This approach reflects: This development philosophy values clear documentation, readable code, and belief that good architecture emerges from simplicity rather than being imposed through complexity.
Exceptions: - Auto-generated files (.g.cs, .designer.cs, files with <auto-generated> markers) - Files in bin/, obj/ directories - Third-party or vendored code The following three constants must stay in sync. If a PR changes any one of them, the reviewer (human or Copilot) must ask the author to confirm the other two are still correct: Snapshot tests in src/Tests/Microsoft.Agents.A365.Observability.Runtime.Tests/Tracing/Exporters/ExportConfigConsistencyTests.cs will fail if any value drifts, but the developer must also verify the values are correct for the target environment — the…
A file Claude Code reads at the start of every session. It holds the commands, conventions and warnings the agent needs for this project.
Where does it go?
At the repository root. Claude Code also reads CLAUDE.md files in subdirectories when it works there.
What should it contain?
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.md or AGENTS.md?
Claude Code reads CLAUDE.md; most other agents read AGENTS.md. Many projects keep one and point the other at it.
MicrosoftCLAUDE.md