mirror of https://github.com/cloudflare/vinext.git synced 2026-09-14 19:04:59 +08:00

Files

T

Nathan Nguyen 5b5071cd5c ci(bonk): downgrade bots from gpt-5.5 to gpt-5.4 (#900 )

bonk and bigbonk currently throw ProviderInitError on every invocation
after the gpt-5.5 switch in #898. The lookup-side regression that #899
patched was masked by the opencode bump (1.14.22 -> 1.14.25) but a
second failure now surfaces during SDK init for cloudflare-ai-gateway.
opencode swallows the underlying cause, and the gpt-5.5 path through
ai-gateway-provider has no working precedent in this repo, so iterating
on it would block PR review indefinitely.

Roll the model back to gpt-5.4, which has a known-good path through
opencode + ai-gateway-provider + Cloudflare AI Gateway, and remove the
opencode.json workaround introduced in #899. Reasoning effort stays at
"xhigh" per James's preference; opencode will fall back to its default
effort if gpt-5.4 doesn't accept it.

Once gpt-5.5 SDK init is fixed upstream (opencode + ai-gateway-provider
+ models.dev cloudflare-ai-gateway entry), we can roll forward again.

2026-04-25 19:54:48 +00:00

2.6 KiB

Raw Permalink Blame History

Contributing to vinext

vinext was born from an experiment in pushing AI to its limits. Almost every line of code in this repo was written by AI, and every pull request is both created and reviewed by AI agents. We welcome human contributions, but if you want to have a good time in this repo, you're going to want to use AI.

Recommended setup

We use OpenCode with GPT-5.4, set to xhigh reasoning effort. This is the same setup that built the project.

Before you open a PR

Run the test suite. pnpm test runs Vitest. pnpm run test:e2e runs Playwright.
Add tests for new functionality. Unit tests go in tests/*.test.ts. Browser-level tests go in tests/e2e/.
Run the static checks. pnpm run check runs formatting, linting, and type checks via Vite+.
Read AGENTS.md. It has the architecture context, key gotchas, and development workflow that will save you (and your AI) time.

AI code review

Every PR goes through AI code review. When you open a PR, a contributor with write access will request a review from BigBonk (GPT-5.4, xhigh reasoning effort). External contributors can't trigger this directly.

Our process is to iterate on BigBonk's feedback until there are no unresolved comments. That doesn't mean you have to accept every suggestion verbatim, but we've found it to be very good at finding real problems and very useful for debugging this codebase. Expect multiple review rounds on larger PRs.

Our bias is towards merging. This is a new project with known gaps, and we're trying to fill them as fast as possible. We want to get PRs in, not block them. If your contribution moves things forward, we'll work with you to get it landed.

At a maintainer's discretion, we may push BigBonk's recommended changes directly into your PR and merge it, or we may create a new PR based on your work. When we do this, we always try to preserve the original author credit. If that ever doesn't happen, please let us know and we'll fix it.

Debugging

For browser-level debugging (verifying rendered output, client-side navigation, hydration behavior), we recommend agent-browser. Unit tests miss a lot of subtle browser issues. agent-browser has been effective at catching them throughout this project.

What to work on

Check the open issues. If you're looking to contribute, those are a good place to start.

Project structure

See AGENTS.md for the full project structure, key files, architecture patterns, and development workflow.

2.6 KiB Raw Permalink Blame History