ensemble-review

name	ensemble-review
description	Review code changes using multiple models in parallel, then synthesize a unified report.

Ensemble Review

Steps

1. Get the diff to review:

   • Run git diff main...HEAD via bash to get the current branch's changes
   • If no changes found, tell the user there are no changes to review

2. Spawn 3 subagents, each with a different model. Give each the diff as part of the prompt:

   Models to use:

   • oci/openai.gpt-5.4
   • oci/google.gemini-2.5-flash
   • oci/xai.grok-code-fast-1

   Use spawn_subagent for each model with this prompt (include the diff inline):

   Review this code diff. Identify bugs, security issues, performance problems, and readability concerns. For each issue, provide the file, line, severity (critical/warning/info), and a brief explanation.

   {paste the diff here}

   Use tools: ["read_file", "list_files", "grep"] so reviewers can check surrounding code.

3. Collect all 3 reviews. Note any subagents that failed and which models succeeded.

4. Synthesize a unified report:

   • Consensus issues: flagged by 2+ models (high confidence)
   • Unique findings: flagged by only 1 model (review manually)
   • Summary: overall assessment combining all perspectives

5. Present the report to the user.