AI Test Case Generator

01. Project Overview

This is the flagship Installable Quality Engineering product on this site: turn real requirement artifacts into reviewable QA deliverables. Generation always runs a human review loop with Test Manager · Developer Manager · Product Manager roles, scoring coverage / logic / executability / clarity / security (100 points; default threshold 90; max 5 rounds). Official docs: https://innonestx.github.io/testcase-generator/

02. Problem

Why build this?

Real requirements arrive as PDFs, Word specs, screenshots, and short videos. Hand-building scenarios and boundary cases is slow; raw LLM dumps lack coverage, executability, and security perspective — so QA teams cannot adopt them as-is.

03. Solution

How was it solved?

I built an installable multimodal generator: parse PDF / Word / TXT / image / video → AI-generate structured cases → run a structured human review loop (Test / Developer / Product Manager; multi-model config) → iterate with the scoring algorithm until threshold or max rounds → export Excel / Markdown / XMind. The same product supports four run modes: Docker, local source, npm global install, and OpenClaw plugin.

04. Architecture

System architecture

05. Key Features

Core capabilities

• Multimodal intake: PDF, Word, TXT, images, and video (optional ffmpeg for frames)
• Human review loop: Test Manager (coverage/executability) · Developer Manager (feasibility/security) · Product Manager (business/journey)
• Scoring dimensions: Coverage 30 · Logic 20 · Executability 20 · Clarity 15 · Security 15
• Multi-vendor model slots: Anthropic, OpenAI, DeepSeek, Qwen, Gemini, or custom OpenAI-compatible endpoints
• Export to Excel, Markdown, and XMind; English and Chinese generation
• Four install surfaces: Docker, local source, npm, OpenClaw plugin

06. My Role

What did I specifically do?

• Owned product direction: installable quality tool, not a one-off demo notebook.
• Designed the multimodal parse → AI generation → scored human review → export pipeline.
• Shipped four run paths: Docker, local source, npm global, and OpenClaw plugin.
• Wired review thresholds, max rounds, and QA-native export formats into the default runnable config.

07. Tech Stack

What technologies power it?

• TypeScript
• Express (Standalone Web)
• Anthropic / OpenAI-compatible LLMs
• Docker
• npm
• OpenClaw Plugin
• Excel / Markdown / XMind export

08. Metrics

Docker pulls refresh live via the Docker Hub API Gateway; npm and OpenClaw / ClawHub stay in sync.

09. Deployment

How do you run it?

Distribution centers on Docker, npm, and OpenClaw (local source is also supported). Docker pulls are collected live via the Docker Hub API Gateway.

1. Docker: docker pull xuxuclassmate/testcase-generator:latest, set AI_PROVIDER + API key, map 3456, open http://localhost:3456.
2. npm: npm install -g @classmatexuxu/testcase-generator, then testcase-generator --standalone.
3. OpenClaw: openclaw plugins install -l /path/to/testcase-generator, configure models[] in config.yaml, use /testgen.
4. Local source (no download counter): clone → npm install → configure .env → npm run build && npm run start.

10. GitHub

11. Live Demo

12. Tradeoffs

• Chose a scored human review loop (Test / Developer / Product Manager) over raw LLM dumps — coverage, executability, and security are first-class defaults.
• Shipped Docker + source + npm + OpenClaw together so humans and agents share one product (Installable > Demoable).
• Accepted multi-round review latency in exchange for Excel / Markdown / XMind artifacts QA teams already use.

13. What shipped

• Flagship installable AI testing product on this portfolio
• Four official run modes: Docker · source · npm · OpenClaw
• Human review + score threshold baked into the default runnable config
• Live proof via Docker pulls, npm downloads, and ClawHub installs

14. Inspectable proof

15. Links

Install paths, proof, and deeper reading.