LaunchKit · 2026
Back to Skills

cross-model-review

Adversarial plan review using two different AI models.

0
431 downloads
by @don-gbot

Setup & Installation

openclaw skills install @don-gbot/cross-model-review

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install cross-model-review

Version History

v2.1.0Mar 5, 2026 - Version note by skill developer:

Round 0 criteria negotiation: Model A proposes 5 task-specific acceptance criteria, Model B challenges/refines. Agreed criteria injected into all reviewer prompts. New command: save-criteria. Backward compatible.

What This Skill Does

Runs adversarial plan review using two AI models from different providers. In alternating mode, each model swaps between writing and reviewing each round, with the loop running fully autonomously until both converge on an approved plan. Intended for plans touching auth, payments, or data models, or any implementation expected to take over an hour.

Unlike single-model review, each model must rewrite based on its own critique, so issues get resolved rather than just listed.

When to use it

  • Reviewing an OAuth integration plan before writing any code
  • Stress-testing a payments flow design against a second model
  • Catching schema migration risks before a database refactor
  • Running autonomous adversarial review before dispatching a coding agent
  • Validating a multi-step feature plan where two engineers would normally disagree

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User asks: Review this plan for adding OAuth2 login to our app

AGENT
  1. 1Initialize a review workspace with the plan file, alternating mode, and two models from different providers
  2. 2Call next-step to get the first action, then spawn Model B as a reviewer sub-agent with the generated prompt
  3. 3Parse the reviewer response into the workspace issue tracker using parse-round
  4. 4Spawn Model B as a writer to revise the plan based on its own review, then call save-plan to store the new version
  5. 5Continue the loop, swapping roles each round, until a reviewer responds APPROVED with no open critical blockers, then call finalize
OUTPUT

plan-final.md with a changelog, rubric scores per round, and a summary of all issues found and resolved

Requirements

Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.

Anthropic API key for Claude modelsOpenAI API key for OpenAI models