Skip to content

feat(evals): add bare React Native quickstart eval - #166

Draft
subhankarmaiti wants to merge 3 commits into
mainfrom
feat/bare-rn-quickstart-eval
Draft

feat(evals): add bare React Native quickstart eval#166
subhankarmaiti wants to merge 3 commits into
mainfrom
feat/bare-rn-quickstart-eval

Conversation

@subhankarmaiti

@subhankarmaiti subhankarmaiti commented Jul 31, 2026

Copy link
Copy Markdown
Contributor

Adds a react_native_quickstart eval covering Auth0 integration in a bare (non-Expo) React Native app, complementing the existing Expo eval. The scaffold is a minimal RN 0.86.2 project — Barkbook login/logout stub, full android/ and ios/ projects, no test or lint tooling — verified to typecheck and bundle.

What it exposes: 20 graders across L1–L5, including L4 checks that the agent wired the native callback config for both platforms — auth0Domain/auth0Scheme manifestPlaceholders in android/app/build.gradle and CFBundleURLTypes in Info.plist — plus L2 checks that it didn't reach for the Expo config plugin or react-native-app-auth.

The Android graders are file-anchored rather than plain substring matches. matches runs against every workspace file concatenated into // FILE: <path> blocks, so an unanchored manifestPlaceholders[\s\S]*?auth0Domain can straddle two files and pass on code that never touched build.gradle:

// FILE: android[\\/]app[\\/]build\.gradle\b(?:(?!// FILE:)[\s\S])*?

@coderabbitai

coderabbitai Bot commented Jul 31, 2026

Copy link
Copy Markdown

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 298219c2-82f2-4e7c-b260-351919f9cc32

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

The Expo and bare RN scaffolds now ship an AGENTS.md stating that no native
toolchain is available and that npm run typecheck is the only verification
needed. The framework renames it to whichever context file the active runner
reads, so all four runners get it.

Environment facts only — SDK API guidance stays in the skills so the no-skills
configs keep measuring what the model knows on its own.
Two bugs made judge graders fail for reasons unrelated to agent output.

The workspace walker reads every file as UTF-8, so the bare React Native
scaffold fed the judge gradle-wrapper.jar and the launcher PNGs as mojibake —
115KB of a 172KB corpus, 5x over maxCodeChars. Every judge grader in that eval
errored without the model being asked. Exclude binaries and generated native
project files; .plist, .xml and .gradle stay in, since judges assert on
Info.plist, AndroidManifest.xml and app/build.gradle. RN is now ~24K chars.

Separately, an oversized corpus threw. That turns an infrastructure limit into a
grader failure, so truncate with a visible notice instead.

Also stop excluding .txt. Baseline mode writes extracted code to
llm_response.txt, so the exclusion left every baseline judge grading an empty
corpus and reporting "the code block is empty" for correct solutions.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant