shenli

shenli / distributed-system-testing

Public

AI-agent skills for distributed-systems testing

agent-skills ai-agents chaos-engineering distributed-systems testing
78
5
89% credibility
Found May 20, 2026 at 108 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

This is a testing toolkit for AI coding assistants that helps teams design and run thorough tests for distributed systems like databases, caches, and distributed services. Instead of writing a few basic tests, it guides your AI assistant to create a structured test plan based on what your system claims to do, then execute that plan while capturing evidence of what actually happens. The result is a detailed findings report with clear verdicts that help you decide whether your system is ready to ship. It works with popular AI coding tools and includes reference materials from academic research on testing distributed systems.

How It Works

1
💡 You discover a smarter way to test

You learn about a testing approach that goes beyond basic tests to find the tricky bugs that distributed systems hide until production.

2
🔌 You connect it to your AI assistant

You paste a simple command to link these testing skills into whatever AI coding tool you already use.

3
📋 You ask for a test plan

You tell your assistant to design a test plan for your system, and it reads your code to understand what your product promises to do.

4
👀 You review the plan together

Your assistant shows you a structured plan with test scenarios, each one designed to prove or disprove what your system claims to do.

5
You run the tests
🔧
Default mode

Your assistant reads your existing tests and tools without changing anything

✏️
Author mode

Your assistant writes test skeletons for you to review before running

🎯 You get clear verdicts

Instead of just pass or fail, you receive a detailed report that tells you exactly what was tested, what broke, and where the problem lives.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 108 to 78 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is distributed-system-testing?

This is a collection of AI agent skills that help you design and run structured tests for distributed and stateful systems. Instead of writing ad-hoc integration tests, you feed your system claims to an AI agent, which generates a formal test plan tied to those claims, then executes it and produces a findings report with 9-state verdicts. The workflow is claim-driven: every scenario is named after the specific claim it tries to falsify, not the setup. For consistency-critical scenarios like durability or linearizability, the plan enforces a model/history/checker discipline with named checkers like Porcupine or Elle. The output is plain Markdown artifacts that a human reviewer can read and make a ship/no-ship call without re-running anything.

Why is it gaining traction?

The hook is enforcing rigor where most teams just wing it. Distributed systems fail in production in ways integration tests rarely catch: partial partitions, crash-recovery bugs, idempotency under replay. This project gives AI agents an opinionated workflow pulling from Jepsen, Elle, and years of distributed systems testing literature. The 9-state verdict taxonomy means "the chaos script ran cleanly" cannot be mistaken for "the claim survived." Every FAIL gets tagged with SUT/harness/checker/environment blame so bugs reach the right queue. Works with any agent that reads Markdown and runs shell, which covers Claude Code, Copilot CLI, Cursor, Codex, and Gemini.

Who should use this?

Backend engineers building distributed databases, message queues, or consensus systems who want more than happy-path coverage. Platform teams validating that their fault-tolerance claims actually hold. Anyone maintaining systems where partial failures, crash recovery, or ordering guarantees are load-bearing. Not for teams still iterating on core functionality; the methodology assumes you have claims worth testing.

Verdict

With 78 stars and a credibility score of 0.9%, this is early-stage but grounded in serious distributed systems literature. The verification directory shows real runs against AgentDB with concrete findings surfaced. If you're building or maintaining distributed infrastructure, the claim-driven workflow and verdict taxonomy fill a gap that ad-hoc testing leaves wide open. Worth evaluating now, but treat it as a methodology toolkit that will evolve.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.