moai-workflow-testing

Characterization tests capture what the code currently does, not what it should do. Follow the PRESERVE phase: write tests documenting existing behavior, record input-output snapshots for complex scenarios, verify all tests pass, then refactor while keeping them green.

TRUST 5 is a quality assessment model scoring code across five dimensions: Testability, Readability, Understandability, Security, and Transparency. Scores are combined with weighted averaging, and critical issues in any dimension can override the average to block progression.

A Haiku agent first filters out closed, draft, or trivial PRs. Then five parallel Sonnet agents check CLAUDE.md compliance, obvious bugs, git history, previous comments, and comment accuracy. Issues scoring below 80 confidence are filtered before posting via the gh CLI.

The workflow supports Python with pytest, JavaScript and TypeScript with Jest or Vitest, Go with go test, Rust with cargo test, plus Elixir, PHP, and Ruby toolchains. Each language pairs with matching static analysis and security scanning tools.

Default quality gates require an 85 percent coverage threshold for changed packages, a minimum TRUST score around 0.85, and zero critical issues. Gates can run in strict, standard, or lenient modes depending on project requirements.

Environment differences between local machines and CI runners cause CI-only failures. The workflow guidance is to trust CI output over local runs, fix flaky tests instead of re-running them, and verify stability with repeated executions such as -count=3.