Code Review

Abbado offers two ways to review agent-generated code: a quick one-shot AI review and an interactive reviewer agent that runs in its own terminal.

One-Shot Review

The simplest option. Click Review in the Changes tab and Abbado sends the current diff to an AI model via claude -p. The response comes back as structured markdown covering:

  • Potential bugs or logic errors
  • Code style and best practice suggestions
  • Security concerns
  • Missing edge cases or error handling

This is fast and requires no configuration — it works on any session out of the box.

Interactive Reviewer Agent

For deeper, conversational review, you can attach a reviewer agent to a session. The reviewer is a second AI agent that gets its own PTY terminal in the Review tab.

How It Works

  1. Configure a reviewer agent at session creation, or add one later via session settings
  2. The reviewer agent launches in its own terminal, independent of the main agent
  3. Click Send Review to pipe the current diff into the reviewer agent's terminal
  4. The reviewer analyzes the diff and responds with feedback
  5. You can continue the conversation — ask follow-up questions, request specific checks, or discuss tradeoffs

Cross-Provider Review

The reviewer agent does not need to use the same provider or model as the main agent. This is one of the most powerful patterns in Abbado:

  • Have Claude Code write the code, then Codex review it
  • Use a larger model for review (e.g., Opus reviewing Sonnet's output)
  • Use a cheaper model for quick sanity checks

Each agent is independently configured with its own provider, model, and instructions.

Configuration

You can set the reviewer agent in two places:

  • At session creation — select a reviewer agent from the dropdown alongside the main agent
  • After creation — update the reviewer via session settings or the API:
PUT /api/sessions/:id/reviewer
Content-Type: application/json

{
  "reviewer_agent_id": "agent-uuid-here"
}

Setting the reviewer agent ID to null removes the reviewer from the session.

Choosing the Right Mode

Feature One-Shot Review Reviewer Agent
Setup required None Create a reviewer agent config
Interactivity Single response Full conversation
Cross-provider No (uses Haiku) Yes (any provider/model)
Best for Quick sanity check Deep review, follow-up questions