How does Coherence Check prevent AI hallucinations?

The skill uses strict context isolation to keep raw data separate from reasoning and requires unanimous approval from independent agents with misaligned failure modes.

What is the Team of Rivals architecture in Coherence Check?

It is a multi-agent system where specialized critics (Code, Security, and Domain) provide independent evaluations with veto power to ensure code meets high reliability standards.

Does using this skill increase token costs?

Yes, it typically introduces a ~38.6% computational overhead due to the multiple agent evaluations, which is a calculated investment to achieve an 80% reduction in user-facing errors.

When is it best to use the Coherence Check skill?

Use it for high-stakes code such as financial calculations, authentication flows, and database migrations where errors have significant business or security consequences.

Can I customize the verification critics?

Yes, you can configure which critics are enabled, select specific models, and define domain specializations like financial, healthcare, or legal in your local configuration files.

Coherence Check

Name: Coherence Check
Author: reggiechan74

byreggiechan74

•

Security & Testing

Implements a multi-agent "Team of Rivals" verification workflow to ensure production-grade reliability for critical code changes.

Coherence Check is a high-reliability verification skill designed for production-critical operations like financial calculations, security implementations, and database migrations. By orchestrating specialized Code, Security, and Domain critics with opposing incentives, it achieves a 92%+ success rate in catching errors that single-agent models often miss. The skill enforces a strict pre-execution planning phase, context isolation to prevent hallucinations, and a consensus-based retry loop, making it an essential tool for developers working on high-stakes codebases where errors carry significant consequences.

Key Features

01Automated audit trails and decision logs for full traceability of code changes

023 GitHub stars

03Strict context isolation to prevent hallucinations and maintain data integrity

04Hierarchical veto authority requiring unanimous approval for high-stakes deployments

05Multi-agent "Team of Rivals" architecture featuring Code, Security, and Domain critics

06Pre-declared acceptance criteria and automated success gating

Use Cases

01Auditing authentication flows and security-sensitive implementations against OWASP standards

02Verifying complex financial logic and payment processing systems for accuracy

03Validating high-risk database schema changes and mission-critical production migrations

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add reggiechan74/cc-plugins coherence-check

For use in Claude.ai and ChatGPT

Download Skill