Skip to content
llm-review-panel

Quick start

This walks through a real run with the default panel: Claude Code and OpenCode against a local Ollama model. Adjust to whatever reviewers you have enabled.

  1. Make sure the reviewers are reachable.

    Terminal window
    bin/llm-review-panel review docs/example-plan.md --dry-run

    Confirm both claude and opencode print found on PATH. Fix any missing binaries before continuing.

  2. Write a plan to review.

    Drop your plan into any markdown file. For example:

    # Plan: extract auth middleware
    

    Move the inline auth check from `src/Controller/*.php` into a shared
    middleware. Wire it up via the kernel and add a smoke test.

  3. Run the review.

    Terminal window
    bin/llm-review-panel review path/to/plan.md

    You’ll hit three checkpoints:

    • Checkpoint 1 confirms the assembled prompt and lists the enabled reviewers (with a warning if any are paid).
    • Checkpoint 2 shows each raw review. Continue, re-run a single reviewer by id, re-run all failed reviewers at once (offered when any failed), or abort.
    • Checkpoint 3 shows the consolidated synthesis. Accept and persist, or abort.

  4. Inspect the persisted run.

    On accept, every artifact lands under .aireview/runs/<timestamp>/:

    .aireview/runs/2026-06-17T14-32-08/
    ├── prompt.md
    ├── reviews/
    │   ├── claude.json
    │   ├── claude.stderr.log
    │   ├── opencode.txt
    │   └── opencode.stderr.log
    ├── synthesis.md
    └── manifest.json

    manifest.json carries timings, statuses, and a snapshot of the config that produced this run.