skills.rest
brunoldqueiroz
brunoldqueirozCOMMUNITY
Explore all skills by this author →
GitHub

self-consistency

Generate, score, and select the best candidate solution among parallel agent options.

Updated Aug 23, 2026
Search Tags:#decision making#code generation#validation#evaluation#orchestration#candidate selection
One-click install
npx skills add https://github.com/brunoldqueiroz/marvin --skill self-consistency
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: self-consistency
Source: https://github.com/brunoldqueiroz/marvin/tree/main/.claude/skills/self-consistency
Command: npx skills add https://github.com/brunoldqueiroz/marvin --skill self-consistency

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill tackles high-stakes decision-making by generating multiple candidate solutions, objectively scoring them against defined criteria, and selecting the most confident winner, ensuring robust and validated outcomes.

Core Features & Use Cases

  • Parallel Candidate Generation: Creates three distinct solutions simultaneously, each optimized for a different stance (simplicity, performance, extensibility).
  • Weighted Rubric Scoring: Evaluates candidates against a customizable rubric, ensuring objective comparison.
  • Confidence-Based Selection: Determines the winner's confidence level based on score spread, flagging close calls for user escalation.
  • Use Case: When deciding between different architectural approaches for a new feature, this Skill can generate three distinct designs, score them on correctness, maintainability, and performance, and present a clear, data-backed recommendation.

Quick Start

Use the self-consistency skill to compare three different implementations for the new user authentication module.

Frequently Asked Questions about self-consistency

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

FAQPage Schema
How do I validate and select the best architectural design from multiple options?▼

Validating architectural designs involves generating multiple parallel solutions, scoring them against a weighted rubric for correctness and maintainability, and selecting the winner using explicit confidence scoring.

What is a self-consistency workflow for complex decision-making?▼

A self-consistency workflow orchestrates parallel agent execution to generate distinct options, evaluates them using weighted rubrics, and selects the most confident candidate to ensure robust, validated outcomes.

How do I score candidate solutions objectively during code generation?▼

Score candidate solutions objectively by applying a customizable weighted rubric that evaluates each parallel-generated option against defined criteria like simplicity, performance, and extensibility.

Can I use parallel agent execution for evaluating different code implementations?▼

Yes, parallel agent execution evaluates different code implementations by generating three distinct solutions simultaneously, each optimized for a different stance, and scoring them against defined criteria.

What happens when confidence scores are too close during candidate selection?▼

When confidence scores are too close during candidate selection, the system determines the winner's confidence level based on the score spread and flags the close call for user escalation.

When should I not use weighted rubric evaluation for decision-making?▼

Avoid weighted rubric evaluation for decision-making when dealing with simple tasks that do not require generating multiple parallel candidate solutions or when robust validation and memory persistence for prior evaluations are unnecessary.

Related Skills

Explore More Software Engineering Skills→
Ankh-StudioAnkh-Studio

agent

Evaluate AI agent artifacts against a predefined rubric with weighted scores.

Official
Intermediate
juanmanueldazajuanmanueldaza

sdd-propose

Generate and evaluate 2–3 approaches for planned software changes.

Community
Intermediate
ai-pivotai-pivot
12

worktree

Create isolated git worktrees for parallel agent workflows and coordinated merges.

Official
Advanced
rm2thaddeusrm2thaddeus
4

ensemble-solving

Generate multiple diverse solutions and select the best via weighted rubric.

Community
Advanced