Suggestions

Generate suggestions to find potential issues with generated code.

What are suggestions?

Suggestions are actionable recommendations that help catch issues in code generated by Agents.

Each suggestion includes:

  • A title summarizing the issue

  • A description explaining the problem with more detail

  • An action that can be taken to fix the issue

[!TIP] Mental model Think of the Verifier as an LLM-based code reviewer that evaluates the agent's work at the end of each turn. It audits both the code changes and the conversation history to verify the agent's work matches what it claimed and meets quality standards, generating suggestions based on issues it identifies.

How suggestions are generated

  1. After an agent finishes its work, the Verifier reviews what happened, looking at the code changes and what the agent said in the conversation.

  2. It compares this against what you originally asked for, flagging anything that looks off: test failures, type errors, logic issues, missing pieces, and so on.

  3. When the Verifier finds problems, it turns them into clear suggestions you can act on.

Depending on your settings, this can run automatically after each agent turn or be triggered manually.

Last updated