Use case

AI Agents for QA Testing

Name: Fleet
Author: Fleet

QA is often the last step before release and the first to be compressed when a deadline moves. Manual QA at scale requires a QA engineer to re-execute the same test scenarios repeatedly across browsers, environments, and edge cases. Even with good automation coverage, someone has to write the tests, maintain them as the product changes, and triage failures.

The result is that QA either becomes a bottleneck or a rubber stamp. Neither is useful. Teams need QA that scales with development velocity without requiring proportional headcount.

How it works with an agent fleet

A qa-engineer agent is dispatched when a PR is ready for review. It runs the existing test suite, checks coverage reports, and uses Claude Code to inspect untested code paths. It publishes a structured result to fabric.

# .fleet/config.yaml
agents:
  - name: qa-engineer
    role: qa-engineer
    model: claude-sonnet-4-6
    subscribes_to: pr_needs_review

The agent's prompt lives at .fleet/prompts/qa-engineer.md — Fleet resolves it by convention from the agent name, so there is no prompt: field in config. The qa-engineer prompt instructs the agent to run tests, identify gaps, and write additional tests for uncovered paths. It uses the /fleet-review-pr skill which handles checkout, test execution, and structured feedback.

The fleet pattern

QA agent runs on every PR in the review chain. If tests fail, it publishes pr_changes_requested with the failure details. The developer agent picks up that event, fixes the failure, and the chain restarts. No human coordination required for the mechanical loop.

Guardrails that matter here

Run-time (duration) budget on the qa-engineer prevents extended test-writing sessions without human review of the new tests
QA agent cannot merge — it only publishes fabric events that other agents and humans consume
All test additions are committed to the PR branch and go through normal code review before merge

Who this is for

Engineering teams shipping frequently who want QA to run on every PR without manual coordination. Also useful for teams that have coverage gaps and want an agent to identify and fill them incrementally.

Frequently asked questions

Can the QA agent test frontend applications?

The agent can run any test command you configure in the prompt, including Playwright, Cypress, or any browser automation tool installed in the environment. It does not have a built-in browser — it delegates to your existing test tooling.

What if the agent writes bad tests?

New tests are committed to the PR branch and reviewed by the tech-lead agent and any human reviewers before merge. The qa-engineer does not have direct push access to your main branch.

Related use cases

AI Agents for Code Review →AI Agents for Test Generation →AI Agents for Bug Fixing →

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.

See how it works Install Fleet