Independent review before your agent acts.
Rejudge
Fresh · read-only
does this migration need a lock?
gpt-5.5 (judge) 0. thinking 01s weighing the two objections
⎿ deepseek-v4-pro ✓ done (35s | 6 tools)
⎿ mimo-v2.5-pro ✓ done (41s | 4 tools)
⎿ gpt-5.4 2. read 03s db/migrate/0142_add_index.sql
Total 52s
Why
The agent that just wrote the code is the worst candidate to review it. It checks its own reasoning with the model that produced it, so a mistake made while writing survives the reading.
Asking one other model helps — until that model is confidently wrong. You get a fluent second opinion with nothing to weigh it against, and no signal about which parts are actually contested.
Rejudge runs several models that never see each other's work, then makes one of them account for the disagreement. What comes back is either agreement worth trusting or a dissent worth reading.
How it works
Install
Node 22.19 or newer. The package ships prebuilt, so Bun is needed only to build from source.
npm install -g rejudge
Create .rejudge/config.json in the project you want reviewed. Two reviewers minimum, and every model carries a reasoning level —
{
"reviewers": [
"opencode-go/deepseek-v4-pro@high",
"opencode-go/mimo-v2.5-pro@high",
"opencode-go/minimax-m3@high"
],
"judge": "opencode-go/glm-5.1@high"
}
A project config wins over the global one at ~/.config/rejudge/config.json. Credentials stay in the environment; the CLI never stores a key.
rejudge "does this migration need a lock?"
git diff | rejudge "review this change"
rejudge --resume <run-id> "what about the rollback path?"
The answer owns stdout; progress, configuration, and the run id go to stderr, so redirecting the answer keeps it clean. Inside Pi the same package registers a native rejudge tool plus the /rejudge and /rejudge-diff workflows.