Summary

Adds a CI workflow that builds and runs the unit tests on every push and on pull requests targeting main, and hardens the test suite against a timing-sensitive flaky failure so CI stays reliably green/red for the right reasons.

Changes

CI workflow (.github/workflows/assembly.yml)

• Run on push (all branches) and pull_request targeting main, so PRs get validated before merge.
• Add least-privilege permissions: contents: read.
• Add a concurrency group keyed on github.ref with cancel-in-progress: true so superseded runs on a branch/PR are cancelled.
• Enable Gradle caching in setup-java and build with ./gradlew build --stacktrace for clearer failure diagnostics.
• Publish a JUnit test-results summary via dorny/test-reporter@v3 (native Node.js 24 runtime). fail-on-error: true so the published check turns red when any test fails instead of masking failures.

Flaky test mitigation

The robot battle tests are timing-sensitive and can intermittently report transient engine errors (e.g. Unable to stop thread) on slow/loaded CI machines. Addressed on two complementary fronts:

1. Engine resilience (RobotThreadManager): make robot thread shutdown tolerant of CPU starvation rather than assuming a stuck thread. Extract the magic timeouts into named constants, give the shutdown path a larger overall budget, re-assert the interrupt periodically (a single early interrupt can be missed before the thread reaches an interruptible point), and Thread.yield() while polling so a contended target thread gets scheduled.
2. Gradle test-retry plugin (CI safety net): retry failed tests a couple of times via a shared convention so genuine flakiness does not fail the build, while real failures (all retries exhausted) still fail. Wired through gradle/libs.versions.toml, buildSrc, and the java-conventions convention plugin (gradlePluginPortal() added for plugin resolution).

Notes

• test-retry only masks eventually-passing flakes; a test that fails all retries still fails the build and the test-reporter check.