How does this skill prevent AI hallucinations in code?

It implements a 'No Fabrication' rule that forbids claiming code is functional or tests have passed unless the agent has actually executed the commands and verified the output.

Can this skill help with project-specific coding standards?

Yes, Rule 4 (No Invented Conventions) requires the agent to cite specific source files like .eslintrc or CLAUDE.md before asserting that a project convention exists.

What is the mandatory 'What's not covered' section?

Every implementation must end with a list of known limitations, unhandled edge cases, and risks to ensure the developer is fully aware of the solution's scope and boundaries.

Does it impact the speed of code generation?

While it adds a pre-delivery checklist, it significantly reduces debugging time by catching 'reasoning drift' and silent errors before the code is delivered to the user.

Implementation Integrity Guard

Name: Implementation Integrity Guard
Author: renathoaz

byrenathoaz

•

Developer Tools

Enforces strict behavioral constraints and transparency standards to ensure Claude produces honest, verified, and high-integrity code implementations.

The Guarding skill transforms Claude into a disciplined engineering partner by applying rigorous integrity guardrails to the implementation process. It specifically targets common AI failure modes such as 'reasoning drift,' where code contradicts previous analysis, and 'optimistic fabrication,' where results are claimed without actual execution. By mandating a standardized 'What's not covered' report and requiring traceable citations for project conventions, this skill ensures that every deliverable is transparent about its limitations and verified for accuracy, making it essential for production-grade development.

Key Features

01Reasoning-Action Coherence: Validates that final code reflects all concerns raised during the analysis phase.

02Verified Convention Citations: Ensures project rules are cited directly from config files rather than assumed.

03Anti-Fabrication Protocol: Prohibits claiming 'tests pass' or 'code works' without recorded execution or verification.

042 GitHub stars

05Confidence Indicators: Labels outputs with High, Medium, or Low confidence based on the level of verification performed.

06Mandatory Transparency Reporting: Forces the inclusion of a 'What's not covered' section to highlight edge cases and risks.

Use Cases

01Preventing silent omissions of edge cases during complex refactoring tasks.

02Maintaining strict adherence to project-specific style guides and linting rules during code generation.

03Enforcing honest reporting when third-party API mocks or stubs are used instead of real logic.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add renathoaz/arsyn guarding

For use in Claude.ai and ChatGPT