How does Team Spec Refinement differ from standard refinement?

Unlike sequential pipelines where each agent works in isolation, this skill uses a live debate format where the Advocate can push back on the Proposer in real-time, creating richer adversarial tension and better output.

When should I use this skill instead of simpler methods?

This skill is specifically designed for Large (L) and Extra-Large (XL) complexity tasks. For small or medium tasks, simpler workflows like dm-work:dialectical-refinement are recommended.

What are 'Too Thin' indicators?

These are automated checks used by the Lead agent to ensure the scope hasn't been cut too aggressively. For example, it triggers if a system has fewer than 5 features or if structured output capabilities are removed.

Which AI models are used in the team composition?

The skill utilizes a mix of models: Haiku for rapid initial analysis and Opus for the Proposer, Advocate, and Lead roles to ensure high-reasoning capabilities during the debate.

Team Spec Refinement

Name: Team Spec Refinement
Author: rbergman

byrbergman

•

Productivity & Workflow

Transforms ambiguous software specifications into actionable work items through live, multi-agent adversarial debate.

Team Spec Refinement is a sophisticated capability for Claude Code designed to handle large-scale, high-complexity tasks. By orchestrating a team of specialized AI agents—including an Analyst, Proposer, Advocate, and Judge—it moves beyond sequential processing to create a live debate environment. This adversarial approach ensures that requirements are stress-tested, ambiguities are surfaced early, and scope is locked down with high-quality acceptance criteria before a single line of code is written, making it ideal for large (L) and extra-large (XL) software projects.

Key Features

01Structured output compatible with dialectical-refinement standards

02Automated 'Too Thin' indicators to prevent excessive scope reduction

03Defined scope-locking phase to ensure architectural integrity

04Live multi-agent adversarial debate between specialized AI roles

052 GitHub stars

06Multi-model optimization using Haiku for analysis and Opus for reasoning

Use Cases

01Converting high-level project 'beads' or vague ideas into technical specifications

02Identifying hidden dependencies and gaps in complex feature requests

03Ensuring architectural consistency and feature preservation in large-scale builds

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add rbergman/dark-matter-marketplace refinement

For use in Claude.ai and ChatGPT

Key Features

01Structured output compatible with dialectical-refinement standards

02Automated 'Too Thin' indicators to prevent excessive scope reduction

03Defined scope-locking phase to ensure architectural integrity

04Live multi-agent adversarial debate between specialized AI roles

052 GitHub stars

06Multi-model optimization using Haiku for analysis and Opus for reasoning

Use Cases

01Converting high-level project 'beads' or vague ideas into technical specifications

02Identifying hidden dependencies and gaps in complex feature requests

03Ensuring architectural consistency and feature preservation in large-scale builds

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add rbergman/dark-matter-marketplace refinement

For use in Claude.ai and ChatGPT