cross-eval

Reviews board memos across Claude, Codex, and Gemini to reconcile divergent model assessments.

25.3k|3.6k|Updated Oct 19, 2025
One-click install
npx skills add https://github.com/alirezarezvani/claude-skills --skill cross-eval
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cross-eval
Source: https://github.com/alirezarezvani/claude-skills/tree/main/c-level-agents/skills/cross-eval
Command: npx skills add https://github.com/alirezarezvani/claude-skills --skill cross-eval

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

High-stakes decisions like M&A, fundraises, layoffs, and strategic pivots carry single-model bias risk when reviewed by only one AI. This Skill runs the same board memo through multiple model providers and reconciles their disagreements before irreversible commitments.

Core Features & Use Cases

  • Multi-Model Review: Sends the memo to Claude, Codex, and Gemini with an adversarial reviewer prompt, collecting independent votes (APPROVE / REJECT / DEFER) and confidence levels.
  • Graceful Degradation: Detects available API keys and CLIs, falling back to a Claude-only adversarial mode with three distinct prompt personas when other models are unavailable.
  • Structured Consensus Report: Saves a dated Markdown report to ~/.claude/cross-eval/ with vote tallies, consensus and divergent concerns, a GO / PAUSE / STOP recommendation, and open questions for the founder.
  • Use Case: Before signing a term sheet, run /cs:cross-eval on the deal memo to surface risks one model may underweight, such as regulatory exposure or technical scale claims.

Quick Start

Run /cs:cross-eval on my fundraise memo to get independent reviews from Claude, Codex, and Gemini with a reconciled GO or PAUSE recommendation.

Frequently Asked Questions about cross-eval

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get multiple AI models to review a board memo?

Run /cs:cross-eval with the memo path. The command sends the memo to Claude, Codex, and Gemini with an adversarial reviewer prompt, collects each model's top concerns, supports, and vote, then reconciles agreements and divergences into one report.

What happens if only Claude is available for cross-eval?

The command falls back to adversarial single-model mode, running three Claude passes with different system prompts: standard reviewer, devil's advocate, and steelman. The output is clearly labeled as single-model and should be treated as suggestive rather than conclusive.

Does cross-eval require OpenAI or Gemini API keys?

No keys are strictly required. The command probes for OPENAI_API_KEY, GEMINI_API_KEY, or the codex and gemini CLIs and uses whatever is available. Claude alone is sufficient to run, though true multi-model review needs at least one additional provider.

When should I use multi-model review for a decision?

Use it before irreversible, high-cost decisions: signing term sheets, announcing layoffs, entering regulated markets, or any choice where reversal costs more than six months of company time. It is a safety net, not a replacement for outside counsel or a real board.

Where are cross-eval reports saved?

Reports are saved as Markdown files to ~/.claude/cross-eval/ with a date and slug filename. Each report contains the vote tally, consensus and divergent concerns, a GO, PAUSE, or STOP recommendation, and open questions for the founder.