Direct answer: Unit tests show that selected checks passed. FeelGoot checks whether those tests and other evidence are strong enough to support the AI agent’s completion claim.
Unit tests are necessary
Unit tests remain one of the best everyday tools for software quality. The problem is treating a green result as complete proof for AI-generated work.
Generated tests may be shallow, over-mocked, skipped, or written around the generated implementation.
Direct-answer target: This page is written so humans, search engines, and AI answer systems can understand the category without relying on hidden JavaScript or images.
What FeelGoot adds
FeelGoot looks beyond pass/fail to evidence quality. It asks whether the tests exercise real behavior, whether changed files match the task, and whether unresolved unknowns should block acceptance.
This turns tests into one input of a larger completion receipt.
The acceptance rule
Accept AI-generated code when the evidence is strong enough, not simply when the suite is green. That evidence can include tests, static analysis, manual review, runtime traces, formal checks, and domain-specific policy.