Harsh Critic — Completion Bias Defense

To check LLM outputs for completion bias before committing code, you can run automated validation checks against the generated text. This process detects five common completion biases, ensuring declared development tasks are truly complete and preventing omissions or errors.

Common LLM completion biases in development workflows include Completion, Error Loop, QA Skipping, Silent Standard, and Delegation biases. These patterns cause models to declare tasks done prematurely, leading to errors, omissions, and reduced code quality.

Yes, you can automatically validate LLM tool outputs for errors on the fly by auto-running checks after every tool call. This on-the-fly validation applies regex pattern matching to intercept completion biases and ensure code quality during development.

Regex pattern matching detects LLM completion bias by scanning generated outputs for five predefined textual patterns. This automated validation catches specific bias signatures left by the model, ensuring tasks are truly done before integration into development workflows.

Yes, automated completion bias validation integrates with development workflow gates. It connects with the loopy-era-harness plugin to provide further control, applying automated validation at critical stages to maintain code quality and prevent errors.

You should run completion bias checks during LLM code generation before committing code, automatically after every tool call, or on-demand when manually reviewing outputs. This prevents completion biases from introducing errors into your development workflow.