veracidade-confiabilidade

To prevent AI hallucination, enforce a verification-first workflow that checks package files or requests specific file contents before making factual claims, explicitly admitting uncertainty when evidence is missing.

Verify framework versions by requesting the AI to inspect configuration files like package.json instead of guessing, ensuring the assistant reads the actual file contents to confirm the factual version.

Handle AI uncertainty by using clear templates to mark doubt, request user input, or offer verification steps, maintaining a zero-tolerance policy for inventing APIs, file names, or behaviors.

Yes, you can force an AI to admit limitations by requiring it to explicitly mark uncertainty and refuse to produce invented facts when it lacks direct access to sources for verification.

AI invents APIs and file names during code review due to hallucination; applying a zero-tolerance verification rule forces it to search the codebase or request files before making factual claims.

Fact-checking AI responses requires providing the necessary file contents for confirmation, as the assistant must be able to request or read files to verify repository facts instead of guessing.