What is the Deliberate skill for Claude Code?

It is a specialized workflow that uses an adversarial system where subagents are spawned to argue for different options, helping you reach a well-informed decision through rigorous debate.

Does the AI make the final decision automatically?

The skill renders a 'verdict' based on the arguments presented, but it acts as an advisor. You can review the reasoning, trade-offs, and confidence levels before finalizing any choice.

When should I use the adversarial deliberation process?

It is best suited for high-stakes decisions with no single 'correct' answer, such as tool selection, architectural designs, or strategic build vs. buy choices.

How many options can this skill evaluate at once?

The skill requires a minimum of two options to deliberate and has no set maximum, though evaluating 2-4 options simultaneously provides the most clarity.

Adversarial Decision Making

Name: Adversarial Decision Making
Author: chrisallenlane

bychrisallenlane

•

Productivity & Workflow

Facilitates complex decision-making by spawning competing subagents to advocate for different technical or strategic options.

This skill implements a structured adversarial deliberation process within Claude Code to help users navigate difficult choices with significant trade-offs. By spawning dedicated advocate agents for each potential path, the tool ensures every option is rigorously defended and scrutinized through iterative rounds of arguments, rebuttals, and probing questions. This process mimics a legal adversarial system, allowing truth and optimal solutions to emerge from the collision of well-argued positions, ultimately providing a comprehensive verdict that outlines key factors, trade-offs, and confidence levels.

Key Features

01Structured multi-round rebuttal process to surface hidden weaknesses

0212 GitHub stars

03Adaptive questioning mechanism allowing the judge to probe specific claims

04Detailed final judgment reports including confidence levels and trade-off analysis

05Comprehensive fact-finding phase to capture organizational constraints and success criteria

06Automated advocate subagent spawning for parallel option exploration

Use Cases

01Making strategic technical decisions where multiple valid but conflicting approaches exist

02Conducting build vs. buy assessments for third-party tools and libraries

03Evaluating competing architectural patterns or tech stacks for a new project

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add chrisallenlane/claude-swe-workflows deliberate

For use in Claude.ai and ChatGPT

Key Features

01Structured multi-round rebuttal process to surface hidden weaknesses

0212 GitHub stars

03Adaptive questioning mechanism allowing the judge to probe specific claims

04Detailed final judgment reports including confidence levels and trade-off analysis

05Comprehensive fact-finding phase to capture organizational constraints and success criteria

06Automated advocate subagent spawning for parallel option exploration

Use Cases

01Making strategic technical decisions where multiple valid but conflicting approaches exist

02Conducting build vs. buy assessments for third-party tools and libraries

03Evaluating competing architectural patterns or tech stacks for a new project