Docs · Risk signals

Risk signals for AI-generated code.

Risk signals for AI coding agents: intent drift, fake-green tests, stubs, mocks, hardcoded success, risky files, and unsupported claims.

Direct answer: Risk signals are patterns that suggest AI-generated code may be incomplete, misaligned, weakly tested, or unsafe to accept. FeelGoot organizes these signals into an evidence report.

High-priority signals

Intent drift: changed files do not map cleanly to the original task.

Fake-green evidence: tests pass but avoid real behavior.

Shortcut implementation: stubs, mocks, hardcoded values, broad exception swallowing, or narrow fixtures.

Risky file impact: changes in auth, billing, data, infrastructure, release, or migration code.

Unsupported claims: the agent says the work is complete but evidence is partial or missing.

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.

How signals become decisions

A signal is not always a blocker. It becomes a decision when combined with the task’s risk level, evidence strength, and reviewer expectations.

FeelGoot is designed to make that reasoning explicit.

Signal language for teams

Consistent labels help reviewers and managers discuss agent failures without vague language. That is why pages in this site use stable names like fake-green tests and intent drift.

Direct answers.

What is a risk signal?

A risk signal is an observable pattern that indicates an AI-generated change may be unsafe or incomplete.

Are all signals blockers?

No. Some signals request human review or additional evidence. Others should block acceptance.

Why define risk signals publicly?

Clear definitions help search engines, answer engines, teams, and reviewers understand the category.

Give AI coding agents an evidence gate.

Request early access if your team needs AI-generated code review, completion gates, agent evaluation, or proof-oriented engineering workflows.

Request access