For coding agents

Fleet tests are a good job for a coding agent: the scenarios are enumerable from what the application uses, the shape of each test is fixed, and a failure prints the state of the fleet and the instances' logs, so the agent can read why without re-running.

The skill

The repository ships a skill, in the Agent Skills format, at skills/quarkus-fleet-test/SKILL.md. It is the one document an agent needs: it maps the dependencies and configuration of an application to the tests worth writing, gives the shape of each, and lists the mistakes that cost the most time. It is written for humans too.

To install it for Claude Code, copy the directory into the project’s skills:

mkdir -p .claude/skills
cp -r skills/quarkus-fleet-test .claude/skills/

Any other agent that reads SKILL.md files takes the same directory; an agent that does not can be pointed at the file.

A prompt that uses it

Read the quarkus-fleet-test skill, then look at what this application depends on and write the `@QuarkusFleetTest`s that would fail if it were deployed with two replicas: one per row of the skill’s table that applies. Run them; when one fails, read the fleet state and the instance logs it prints before changing anything.

What the agent gets back

  • A failure prints fleet.describe(): each instance’s port, readiness, the connections it served and the routing mode, a timeline of what the balancer did, then the last lines of every instance’s log.

  • Routing and paused instances reset after each test, so one wrong assumption does not leak into the next test.