interpeer

interpeer

Cross-AI peer review for Claude Code.

What this does

Single-model review catches a lot, but models have blind spots: and they tend to have the same blind spots consistently. interpeer sends your code or documents to a different model for a second opinion, with four escalation modes depending on how much scrutiny you need.

Quick (qinterpeer): sends to Codex CLI for fast Claude↔GPT feedback. Takes seconds, catches surface-level disagreements.

Deep (interpeer): sends to Oracle (ChatGPT 5.2 Pro) with prompt optimization and large context support (~200k tokens). Takes minutes, catches architectural and design issues.

Council (winterpeer): full LLM Council with multi-model consensus. Slow, but useful for critical decisions where you want genuine multi-perspective analysis.

Mine (splinterpeer): the interesting one. Instead of just getting a second opinion, this mode extracts the disagreements between models and converts them into actionable artifacts: tests for disputed behavior, specs for ambiguous requirements, and questions for genuinely unclear design choices. Models disagree about interesting things.

Installation

First, add the interagency marketplace (one-time setup):

/plugin marketplace add mistakeknot/interagency-marketplace

Then install the plugin:

/plugin install interpeer

Requires Codex CLI for quick mode and Oracle CLI for deep mode.

Usage

/interpeer

Auto-detects which AI tools are available and offers the appropriate modes.