How does this skill prevent AI hallucinations?

It forces the AI to categorize every statement with certainty labels and cite specific source files or line numbers, ensuring all claims are backed by existing code or documentation.

Can I use this in an English-only project?

Yes. While this localization is for zh-TW users, the skill supports project-specific settings to use English certainty tags by detecting preferences in your CONTRIBUTING.md file.

What are the core certainty labels used?

The core labels include [Confirmed] for direct evidence, [Inferred] for logical deductions, [Assumption] for common patterns, and [Unknown] when information is unavailable.

Does this skill help with architectural decision-making?

Yes, it requires the AI to not just list options, but to provide a specific recommendation with a clear rationale based on the project's current infrastructure and needs.

AI Collaboration Standards (zh-TW)

Name: AI Collaboration Standards (zh-TW)
Author: AsiaOstrich

byAsiaOstrich

•

Collaboration Tools

Standardizes AI interactions through evidence-based certainty labeling and anti-hallucination protocols for software development.

This skill establishes a rigorous framework for AI-human collaboration by requiring Claude to categorize its responses based on verifiable evidence. It utilizes a standardized system of certainty labels—such as [Confirmed], [Inferred], and [Assumption]—to distinguish between direct code facts and logical deductions. By enforcing mandatory source citations (including file paths and line numbers) and providing specific recommendation rationales, this skill significantly reduces AI hallucinations and ensures that developers can trust the technical insights provided during the development lifecycle.

Key Features

01Configurable Tag Language (English or Traditional Chinese via CONTRIBUTING.md)

0240 GitHub stars

03Source Reliability Ranking (Code, Docs, External, Search, Knowledge, User)

04Certainty Labeling System ([Confirmed], [Inferred], [Assumption], [Unknown], [Need Confirmation])

05Mandatory Source Citation (File paths and line numbers)

06Anti-Hallucination Checklists and Protocols

Use Cases

01Ensuring accurate technical onboarding by referencing verified documentation and code snippets.

02Performing code reviews where AI must justify suggestions using specific project patterns.

03Validating complex codebase architecture and dependencies with evidence-backed claims.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add asiaostrich/universal-dev-standards ai-collaboration-standards

For use in Claude.ai and ChatGPT

Download Skill