What the agent does
node-sass is a native addon, and it only works where a prebuilt binding exists for the exact Node ABI. On Node 22, ABI 127, its install script prints Downloading binary from https://github.com/sass/node-sass/releases/download/v7.0.3/linux-x64-127_binding.node and then Cannot download ...: HTTP error 404 Not Found, because the 7.0.3 release shipped Linux bindings for ABIs 72, 83, 93 and 102 only, Node 12 to 17. It then falls back to compiling libsass from source with node-gyp, which fails on any machine without a C++ toolchain, and the package ends up installed without a binding. The first require of it, which for a Create React App project is the first npm start or npm run build, throws Error: Node Sass does not yet support your current environment: Linux 64-bit with Unsupported runtime (127). Even the last release, 9.0.0 from May 2023, stops at ABI 115, Node 20, and npm now marks every version deprecated with the note to use sass or sass-embedded instead.
ibrahimtelman/subcolor, a small React app that generates the colours between two colours, has the usual shape. package.json lists node-sass ^7.0.3 beside react-scripts 5.0.1, react 18.2 and the testing-library packages; yarn.lock resolves node-sass to 7.0.3. Six components import a style.scss next to them, Home.js line 13, Header.js line 8, Color.js line 7, ColorPicker.js line 5, Colors.js line 3 and Copied.js line 4, for 183 lines of SCSS in total, all plain nesting with no @import, no @use and no division. There is no .nvmrc and no CI workflow, so the Node version that builds the site is whatever the developer or the deploy host has, and the deploy script publishes build/ with gh-pages.
The agent first pins the failure to the Node version rather than to the machine. It reads process.versions.modules to get the ABI, runs node-sass's own install script to capture the download URL and the 404, and lists the binding assets of the pinned release and of 9.0.0 from the GitHub releases, so the pull request can show that no node-sass version has ever shipped a binding for this Node. It also tries the install, keeps the node-gyp output, and notes that a successful compile, where a toolchain exists, would only postpone the same problem to the next Node upgrade.
Then it replaces the compiler rather than downgrading Node. It removes node-sass, adds sass, the Dart implementation that react-scripts 5's sass-loader picks up with no configuration, and compiles each of the six style.scss files with the new compiler to check they produce the same CSS; for this repository that is a direct comparison, since nothing in the files depends on libsass-only behaviour. It regenerates yarn.lock, runs yarn build to produce the production bundle and yarn test --watchAll=false for the existing render test, and reads package.json for an engines field or a .nvmrc to add one naming the Node the build was proven on, so the next host cannot silently pick an older runtime.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "This React app fails to start on our current Node with Error: Node Sass does not yet support your current environment, after npm or yarn install printed a 404 for a linux-x64 binding.node download. Read package.json and the lockfile and report the node-sass and react-scripts versions and every file that imports a .scss or .sass file. Pin the cause to the Node ABI: print process.versions.modules, run node-sass's install script and keep the download URL and status it prints, and list the binding assets the pinned release and the newest node-sass release ship, so the pull request shows whether any node-sass version supports this Node. Then replace node-sass with the sass package: remove the old dependency, add sass, confirm the project's bundler or react-scripts picks it up without configuration, and compile every SCSS file with the new compiler, reporting any syntax it rejects (slash division, @import of libsass-only features) with file and line and the smallest change that fixes it. Regenerate the lockfile with the same package manager the repository uses. Run the production build and the existing tests and include their output. Add an engines field or .nvmrc naming the Node version you proved the build on, if the project has neither. Do not change styles, component code or the deploy script. Open a pull request that lists the changes, the versions installed, and what could not be verified without a browser.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
A pull request that changes package.json, the lockfile and, when the project had none, a .nvmrc. Its description opens with the ABI number, the download URL node-sass tried and the 404 it got, and the list of binding assets from the releases page, which together say in one table why no node-sass version was an option. The node-gyp output from the fallback compile sits under it.
The diff to the dependency files is two lines plus the lockfile. A reviewer checks the compile comparison: each SCSS file compiled with the new package, and any line it rejected with the change made. For a repository with plain nested rules that section is empty, and the build and test output below it is the proof that the bundler found the new compiler without configuration.
What the run cannot prove is listed last. The production bundle was built and the render test passed, but no browser loaded the page, so a visual check of the compiled styles, and the gh-pages deploy, are for the reviewer.
Run it on every failed deploy
The break arrives when a host moves its default Node, not when the repository changes, so the alert comes from the deploy. Create an automation on the repository's Project with the prompt above and a github trigger on check_suite.failure; a failed suite's details follow the prompt as event text, and the agent reads the 404 and the runtime number from the log, or answers in one line when the failed suite is not the site build. A project with no CI, like the example, is covered by a schedule trigger with a weekly cron: the agent installs on the current Node and opens a pull request only when the install or the build fails.
A deploy host that posts its own failure notifications usually signs them and cannot add the Authorization: Bearer header a webhook trigger expects, so a handler on your server receives the notification and forwards the body to the trigger's URL with the trigger's secret.
The use case this job follows, with its automation and trigger: Fix a bug from an alert.
Why run it on Rush?
The failure depends on the Node version of the machine, which is exactly what makes it hard to agree on: it builds on the laptop that has Node 16 and fails on the host that has Node 22, and the argument is about environments. The agent runs in a sandbox with the current Node, captures the ABI, the download URL, the 404 and the release assets, and keeps all of it in the session, so the pull request settles the question with four facts instead of a reproduction nobody can share.
Swapping a compiler is a change you want proven file by file, and that is tedious by hand. The agent compiles each SCSS file with the replacement, compares the output, runs the build and the tests, and lists any syntax the new compiler rejected with the line that fixes it. The reviewer reads a diff of two dependency lines and a report, rather than opening every stylesheet.
Frequently asked questions
Can I install a newer node-sass instead?
No release helps. The run lists the binding assets on GitHub: 7.0.3 stops at Node 17 and the last release, 9.0.0 from May 2023, stops at Node 20, so Node 22 has no binding from any version, and npm marks the package deprecated. The sass package is the supported path.
Why did npm install succeed on my laptop but fail on the server?
The laptop has an older Node with a prebuilt binding, or a C++ toolchain that let node-gyp compile libsass. The server has neither, so the install ends without a binding and the first require throws the error in the title. The pull request pins the Node version so both sides build the same way.
Will my SCSS compile the same with sass?
Plain nesting, variables and mixins do. The agent compiles every file with the new package and compares the output; it reports a difference only for syntax libsass accepted and Dart Sass rejects, such as slash division outside math.div, with the file, the line and the change.
Does react-scripts need configuration to use sass?
No. Create React App 5 looks for the sass package first, so removing node-sass and adding sass is the whole change to the toolchain. The run proves it with the production build in the pull request.
Does the agent change my styles or components?
No. The request forbids it. The only edits are in package.json, the lockfile and a file naming the Node version, and any SCSS change is limited to syntax the new compiler rejects, listed line by line.
More jobs like this: Fix a bug from an alert
- Fix "ValueError: 'not' is not a valid parameter name" in a FastAPI App on Python 3.11 or Later (Pydantic)
- Fix "np.alltrue was removed in the NumPy 2.0 release" (NumPy)
- Fix "ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject" (NumPy)
- Fix "AttributeError: module 'numpy' has no attribute 'float'" on NumPy 1.24 and Later (NumPy)
- Fix "AttributeError: module 'numpy' has no attribute 'trapz'" after NumPy 2.4 (NumPy)
- Fix "'onBeforeSetupMiddleware' option is deprecated" from webpack-dev-server (webpack-dev-server)