What are the decision thresholds?

A score of 90% or higher is required to proceed. Scores between 70-89% trigger clarifying questions, while scores below 70% stop implementation to request more context.

Is this skill compatible with my tech stack?

Yes, while it has built-in checks for tools like Supabase and pytest, the framework is designed to verify any technical architecture against existing project standards.

How does it save LLM tokens?

By investing 100-200 tokens in a pre-implementation check, it prevents the model from generating thousands of lines of incorrect or redundant code, saving up to 50,000 tokens per task.

What evaluation criteria does the skill use?

The skill assesses five areas: absence of duplicate implementation, architecture compliance, official documentation verification, OSS reference availability, and root cause identification.

What is the Confidence Check Skill?

It is a specialized framework for Claude Code that evaluates the reliability of a proposed code change before implementation to ensure accuracy and technical alignment.

Confidence Check Framework

Name: Confidence Check Framework
Author: wishdevs

bywishdevs

生産性とワークフロー

Evaluates implementation reliability before execution to prevent architectural errors and maximize token efficiency.

概要

The Confidence Check Skill provides a rigorous evaluation framework for Claude Code, requiring a 90% or higher confidence score before starting any implementation task. By assessing five key criteria—including architectural alignment, documentation verification, and root cause identification—it helps developers avoid redundant work and incorrect technical directions. This preventative approach serves as a high-ROI safeguard, potentially saving up to 50,000 tokens by ensuring the AI has a solid plan and sufficient context before it begins writing code, making it an essential tool for cost-effective and accurate AI-assisted development.

主な機能

ROI-driven token management saving up to 250x the initial check cost
Integration with MCP and WebFetch for real-time documentation verification
Automated duplicate implementation detection using grep and glob patterns
Five-point weighted evaluation system for architectural and logic verification
Decision framework with strict thresholds for proceeding or requesting context
0 GitHub stars

ユースケース

Pre-implementation auditing for complex architectural changes
Resource-conscious development to minimize unnecessary LLM token consumption
Ensuring technical alignment with specific stacks like Supabase, UV, and pytest

概要

主な機能

ROI-driven token management saving up to 250x the initial check cost
Integration with MCP and WebFetch for real-time documentation verification
Automated duplicate implementation detection using grep and glob patterns
Five-point weighted evaluation system for architectural and logic verification
Decision framework with strict thresholds for proceeding or requesting context
0 GitHub stars

ユースケース

Pre-implementation auditing for complex architectural changes
Resource-conscious development to minimize unnecessary LLM token consumption
Ensuring technical alignment with specific stacks like Supabase, UV, and pytest