BuildBuddy

Incremental builds deserve an incremental cache. Major props to Tyler French from the BuildBuddy team for landing this change in Bazel that reduces cache uploads and disk cache size by 40%+ https://lnkd.in/gq4U8Cz8

I'll be at the BUILD Meetup in SF next week, I hope to see some folks there! https://luma.com/gkf54wba

Your coding agent doesn't need a sandbox, it needs a remote execution cluster https://lnkd.in/gZReHcde

Beyond excited to welcome Keith Smiley to the BuildBuddy team! Keith is the maintainer of Bazel's iOS support, a contributor to LLVM and Swift, creator of the Mobile Native Foundation, and an all around awesome human. https://lnkd.in/dEkEtE7A

I'm excited to announce that I joined BuildBuddy! I've been working on bazel as a user of BuildBuddy for a long time, so I'm looking forward to being on this side of things and working with a great team!

Large engineering teams use Bazel to tame their ever-growing monorepos. Today, I'm excited to share rules_codex, a hermetic cross-platform Bazel toolchain and ruleset for OpenAI's Codex coding agent designed to make working in large enterprise codebases fun again. Ever get tired of making sure your docs are up to date and weren't broken by a recent code change? You can write a little codex_test rule to validate that the examples in your README file still work as expected. It's so simple, it fits in a tweet: codex_test( name = "validate_readme", srcs = ["README.mdx"], prompt = "Walk through this README and verify all the steps work correctly.", ) Imagine trying to do this without something like rules_codex. Don't want to write the docs at all? Have Codex write them for you with a codex rule: codex( name = "generate_docs", srcs = ["src/main.py"], out = "docs.mdx" prompt = "Generate markdown documentation for this Python module.") Trying to refactor an old library you're afraid to touch? codex_run( name = "modernize", srcs = glob(["src/**/*.py"]), prompt = "Refactor this code to use modern Python 3.12 features like pattern matching and type hints.") The ruleset is flexible and allows you to run Codex as a build action, in a test, or with bazel run. It also exposes a user friendly toolchain that you can use to write your own genrule, or even a custom ruleset to fit your needs. The best part is, it all works with Bazel's existing infrastructure - so you get the power of remote execution to run 10,000+ prompts sandboxed in parallel with remote execution. Cache results and save tokens with remote caching. Excited to see what you all do with it. My DMs are open for ideas / suggestions / feedback - or feel free to open an issue on the Github repo.

I've been building at the intersection of Bazel + coding agents for a few months and have some fun things to share. The first one is a hermetic, cross-platform Claude Code toolchain let's you easily use Claude in Bazel builds, tests, and runs. Why is this cool? First of all, it lets you use Claude Code in Bazel actions and allows you to take full advantage of Bazel's powerful remote execution, remote caching, and sandboxing features. Remote execution means you can run 10,000+ prompts in parallel in a remote execution cluster like BuildBuddy - each with its own hermetic sandbox and tightly controlled input root for context management. Remote caching means you never have to run the same prompt twice (unless an input changes). This saves you tokens and allows you to build complex pipelines of Claude prompts that only get rerun when needed. Sandboxing (both locally and remotely) means you don't have to worry about Claude Code having access to files that it shouldn't have access to, and you get a clean reproducible sandbox each time. Now for some examples - hate writing docs? Write a Claude rule to generate docs for you: claude( name = "generate_docs", srcs = ["src/main. py"], prompt = "Generate markdown documentation for this Python module.", out = "docs.mdx", ) Want to make sure your README file isn't broken? Write a Claude test to validate that everything still works as expected: claude_test( name = "validate_readme", srcs = ["README.mdx"], prompt = "Walk through this README and verify all the steps work correctly.", ) Or have Claude walk you through a deploy to your favorite cloud provider: claude_run( name = "deploy", srcs = ["main.go"], prompt = "Deploy this app to Google Cloud Run. Ask me for any credentials you need.", interactive = True, ) Want to generate a marketing website for your new Github repo? Point it at the README file and you're off to the races: claude( name = "website", srcs = ["README.mdx"], prompt = "Generate a complete static marketing website based on this README.", ) The possibilities truly are endless here, and I'm really excited to see what you all do with it. Feel free to DM or file an issue on Github with any feedback / ideas / etc you have. What really excites me here is that this takes us one step closer to a world where prompts are the source of truth. By using BUILD files to store and version your prompts in source control - we move from ad hoc prompting to actually engineering prompts and context in a meaningful way. If you're a Codex user, I've got something exciting in store for you tomorrow ;)

Another great technical blog post by Maggie Lou on all the nitty gritty details of how we scale warm runners with snapshotting, chunking, cloning, and more. https://lnkd.in/g-yxX6Xv

Excited to share that Tyler French has joined the BuildBuddy team! He joins us from Uber, and is a maintainer of both rules_go and gazelle. Welcome to the team, Tyler! https://lnkd.in/dCKS7c6A

This is pretty incredible. There was a lot of doubt that we (or any company) could entirely run CI for iOS on Linux, having it use RBE for everything that required MacOS. Brentley Jones proved all the skeptics wrong and it's been a massive improvement to the iOS Developer Experience at Reddit.