What the agent does
The agent reads the failing line and what surrounds it. In a caching gem like seamusabshere/lock_and_cache, lib/lock_and_cache/key.rb builds ALLOWED_IN_KEYS, the set of classes that may appear in a cache key as themselves (lines 44 to 52: String, Symbol, Numeric, TrueClass, FalseClass, NilClass, Integer, Float), and then at lines 54 to 58 reads parts = ::RUBY_VERSION.split('.').map(&:to_i) and unless parts[0] >= 2 and parts[1] >= 4 appends ::Fixnum and ::Bignum. extract_obj_id at line 25 consults the set for every key part. The file is required by lib/lock_and_cache.rb at load, so the constant is evaluated the moment anyone requires the gem.
The check is the bug, and the agent says so with three runs. The condition compares the minor version on its own, so it is true for 2.4 and for 3.4 but false for 3.0 through 3.3. On Ruby 3.1, require 'lock_and_cache' prints key.rb:56: warning: constant ::Fixnum is deprecated and the same for ::Bignum on line 57, and loads. On Ruby 3.3 it raises uninitialized constant Fixnum (NameError) at line 56 and nothing that requires the gem can start. On Ruby 3.4 it loads clean, because the minor number is 4 again. A maintainer who tests on 3.4 would never see it.
The fix is to delete the five lines. Integer is already in the set on line 50, and on every Ruby where the branch would run, an integer is an Integer, so the branch adds nothing a current interpreter needs. The agent then proves it against the gem's own suite, which needs a real Redis: spec/spec_helper.rb opens Redis.new db: 3 for locks and db: 4 for cache, so the agent starts redis-server in the sandbox, installs redis, activesupport, rspec and thread (the gemspec's development dependency bundler ~> 1.6 makes a current bundle install stop and ask for Bundler 1.17.3 instead), and runs rspec on Ruby 3.3, where every example passes. The two somebody already clicking lines in that output come from the spec's own threads at spec/lock_and_cache_spec.rb line 97 and are expected.
It reports the rest without changing it: .travis.yml still lists Ruby 2.3.0 and 2.4.1, the gemspec has no required_ruby_version, and the bundler ~> 1.6 development dependency is what stops bundle install on a current Ruby. Each is one line in the pull request for the maintainer to take or leave.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Our Ruby gem fails to load with \"uninitialized constant Fixnum (NameError)\" on Ruby 3.2 and later. Find every reference to Fixnum and Bignum, explain why any version check around them does not stop the error, and remove or replace them so the code loads on every Ruby from 2.4 up. Show the load on at least one Ruby that warns and one that raises. Run the full spec suite on a current Ruby, starting any service the specs need, and open a pull request with the output and anything else that blocks the gem on current Rubies.",
"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 removes five lines from lib/lock_and_cache/key.rb and changes nothing else. The description shows the require on three Rubies (the warning on 3.1, the NameError on 3.3, the clean load on 3.4) with the version check explained, then the rspec run on Ruby 3.3 passing against a Redis the agent started, and lists the CI file, the missing required_ruby_version and the bundler ~> 1.6 dependency as follow-ups.
A reviewer checks that Integer already covers what the removed lines added, that the suite really ran against Redis rather than being skipped, and decides whether to update the CI Ruby list in the same change.
Run it on every Ruby version bump in a service's Dockerfile or .ruby-version
The error appears the first time an app boots on Ruby 3.2 or 3.3, so the trigger is the change that moves the Ruby version. An automation with a github trigger on pull_request.opened and pull_request.synchronize runs this prompt on each pull request that touches .ruby-version, Gemfile or the Dockerfile, with the event appended after the prompt, and pushes the fix when bundle exec fails to load.
For a fleet of services sharing an in-house gem, your server sends this request once per repository, each with its own UUID as request_id so a retry returns the run already started, and reads each run.completed webhook; a key starts at most max_concurrent runs at once and queues the rest.
The use case this job follows, with its automation and trigger: Fix a bug from an alert.
Why run it on Rush?
A version check that compares the minor number alone is exactly the kind of bug a laptop on the newest Ruby cannot see: it passes on 3.4 and fails on 3.3. In a Rush sandbox the agent runs the load on three interpreters and keeps each output, so the pull request explains the shape of the bug instead of just deleting a line.
The gem's specs need Redis, and a review that skips them proves nothing about cache keys. The agent starts the service, runs the whole suite on a Ruby where the gem could not even load before, and records the count.
Frequently asked questions
Why does it load on Ruby 3.4 but not on 3.3?
The guard reads parts[0] >= 2 and parts[1] >= 4, so it only looks at the minor number. On 3.4 the minor is 4 and the Fixnum lines are skipped; on 3.0 through 3.3 the minor is below 4, the branch runs, and from 3.2 the constant no longer exists. The fix removes the branch rather than repairing the comparison.
Should the lines be replaced with Integer instead of deleted?
No. Integer is already in ALLOWED_IN_KEYS on line 50, so the branch was only ever adding aliases of a class the set already allowed. Deleting it leaves the behaviour on every supported Ruby exactly as it is on 3.4 today.
Why did the agent install gems by hand instead of bundle install?
The gemspec declares a development dependency on bundler ~> 1.6, so a current bundle install stops and asks for Bundler 1.17.3. The agent installs the runtime and test gems directly, runs the specs, and lists the pin as a follow-up so the maintainer can decide how to modernise the gemspec.
Do the specs really need Redis?
Yes. spec/spec_helper.rb connects to two Redis databases at load, so without a server every example errors before it runs. The agent starts redis-server in the sandbox and the pull request shows the full run against it.