Multi-Model Consensus

Runs a high-stakes board memo or strategy brief through multiple AI models (Claude, Codex, Gemini) to get independent reviews and reconcile divergences. It produces a structured cross-eval report with vote tallies, consensus and divergent concerns, and a recommendation, acting as a safety net before irreversible decisions.

✨ What it does

  • Invokes multiple model providers (Claude, Codex, Gemini) with graceful degradation if some are unavailable.
  • Sends each model an independent reviewer prompt that demands top concerns, supports, and a vote (APPROVE/REJECT/DEFER).
  • Reconciles reviews to identify consensus concerns (flagged by ≥2 models) and divergent concerns (flagged by 1 model).
  • Generates a vote tally table and a recommendation (GO/PAUSE/STOP) based on model votes and critical concerns.
  • Lists open questions for the founder arising from divergences.
  • Falls back to adversarial single-model mode (three Claude passes) when only Claude is available, clearly labeling the ou

🎯 When to use it

  • Before signing a term sheet or committing to a major fundraise.
  • Before announcing layoffs or a strategic pivot.
  • Before entering a regulated market or making a regulatory commitment.
  • When the boardroom vote was split or had a critical dissent.
  • For any decision where the cost of reversing exceeds six months of company time.

🚀 How to use

Trigger: /cs:cross-eval

Trigger with the slash command /cs:cross-eval followed by the memo or brief text (or a reference to it). The skill reads the memo, probes for available model CLIs/API keys, and runs the review. Example prompts:

/cs:cross-eval <paste memo here>
/cs:cross-eval @path/to/memo.md

📄 Output: A markdown report saved to ~/.claude/cross-eval/YYYY-MM-DD-<slug>.md containing vote tally, consensus/divergent concerns, recommendation, and open questions.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/c-level-agents/skills/cross-eval .claude/skills/cross-eval

Skill source: c-level-agents/skills/cross-eval/SKILL.md

⚠️ Good to know

Requires API keys or CLIs for Codex and Gemini to achieve true multi-model consensus; otherwise it falls back to a weaker single-model adversarial mode. It is not a replacement for outside counsel or a real board.

❓ FAQ

What happens if only Claude is available?

The command runs in adversarial mode with three independent Claude passes (standard, devil's advocate, steelman) and labels the output as single-model, treating it as suggestive rather than conclusive.

How is the recommendation determined?

It is GO if 2+ models approve and no critical concerns from any model; PAUSE if any model defers or any concern is critical; STOP if 2+ models reject.

What does the output file contain?

It includes the date, memo link, models invoked, a vote tally table, consensus and divergent concerns, consensus supports, a recommendation, and open questions for the founder.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.