cross-critique (warpdotdev/common-skills): a skill that has subagents critique each other's proposals

Warp's skill makes Claude send each subagent's proposal to the other authors for pros and cons, then write one recommendation from the critiques.

Installs: 22K; on skills.sh
Stars: 644; on GitHub
Updated: Oct 2026; last repo commit

The skill is a second round. After several subagents have each written an independent proposal, Claude sends each author the other proposals and asks for pros and cons of each, whether they would change their own answer, and a final ranking with confidence. Claude then writes one answer that leads with a recommendation, and compares critiques by evidence quality, not vote count.

It triggers when proposals diverge on a contested decision: architecture, code review disagreements, design choices or root-cause theories. It says to skip it when proposals already agree or when the question has an objective answer you can check, because the extra round costs time and tokens.

You need the first round of independent proposals first. SKILL.md suggests the council and research skills for that, or any set of subagent proposals you already have.

Good for

Developers weighing architecture or API-shape tradeoffs where agents disagree
Reviewers facing conflicting code review conclusions or competing bug theories
People who already run several subagents and want a sharper synthesis

Install cross-critique

Claude Code

Copy this prompt.

Install the cross-critique skill in Claude Code. Run this command:

npx skills add https://github.com/warpdotdev/common-skills --skill cross-critique

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Run a second round on a contested question by circulating each subagent's independent proposal to the other authors and asking for structured pros and cons, then synthesize. Use this skill whenever you have multiple independent proposals or opinions on a contested decision — architecture tradeoffs, code review disagreements, design choices, competing root-cause theories — and want sharper analysis than you'd produce by synthesizing alone. Pairs naturally with the council and research skills; reach for it liberally whenever proposals diverge.

Sections in SKILL.md

Why this matters
When to use it
Prerequisite: you need independent proposals first
How to do it
1. Assemble the proposals
2. Circulate and ask for structured critique
3. Synthesize
Final answer template
Recommendation
How the critiques shifted things
Why this option wins
Remaining risks and unknowns
Practical notes

Key facts

Author: Warp (warpdotdev)
Repository: warpdotdev/common-skills
Installs on skills.sh: 22,025 as of 2026-11-02
GitHub stars: 644 as of 2026-11-02
Last repo commit: 2026-10-28
Files in the skill folder: 1 file (SKILL.md)
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk pass

What it includes

Collects each author's core recommendation
Keeps proposals short and labels them neutrally, such as Proposal A, B and C. Anonymizing authors is suggested to reduce bias.
1. Assemble the proposals
Sends each author the other proposals
Reuses the same subagents. Asks for pros and cons of each alternative, whether they would revise their own view, and a final ranking with confidence.
2. Circulate and ask for critique
Writes one answer with a recommendation first
Notes where authors converged, the strongest objections, why the winner survives, and remaining risks.
3. Synthesize
Recommendation, shifts, why it wins, risks
A four-section Markdown shape: Recommendation, How the critiques shifted things, Why this option wins, and Remaining risks and unknowns.
Final answer template

From the official docs

Questions

What does the cross-critique skill do?

It runs a second round on a contested question. Subagents that wrote independent proposals review each other's work for pros and cons, then Claude writes one recommendation from those critiques.

How do I install it in Claude Code, the Claude app or Codex?

Use the install box above. The skill is one SKILL.md file. The common-skills README does not document an install method for specific clients.

Does it only work in Warp?

No Warp-specific tool or setting appears in SKILL.md. It only asks the agent to run several subagents, send each one the other proposals and synthesize the replies. It lives in Warp's shared common-skills repo, whose README says the repo holds skills useful across repositories. It needs an agent that can run subagents and send them a second message; the SKILL.md does not name Claude Code or any other client.

Do I need other skills for it?

No, but you need independent proposals first. SKILL.md suggests the council skill for judgment calls and the research skill for investigations, and says any set of subagent proposals works. Those skills are optional companions and are not in this skill's folder.

When should I not use it?

When the proposals already agree, or the question has an objective answer you can verify directly. SKILL.md says critique adds latency and tokens and pays off only when there is real disagreement.

Does it need an API key or scripts?

SKILL.md names no credentials and the folder holds only SKILL.md. It relies on your agent being able to run subagents.

Is it safe to install?

It is one file of instructions and bundles no scripts. SKILL.md says to keep the critique round read-only unless the underlying task involves changes, but it is Claude that follows that. Watch what the subagents you start are allowed to do.

Information

updated Nov 2, 2026

Repository: warpdotdev/common-skills
Folder: .agents/skills/cross-critique

warpdotdev/common-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2

SKILL.md · github.com
common-skills README · github.com