AI QA Agent Platform — Skills-Driven AI Test Automation

A Skills-driven 8-stage automated testing platform. One command to install, natural language to trigger, covering the full QA lifecycle from requirement analysis to regression reporting.

中文

Overview

AI QA Agent Platform breaks down QA workflows into 8 independent Skills, invoked by AI platforms (OpenCode / Claude Code / Cursor) stage by stage. Describe your test goals in natural language, and the platform handles the rest:

qa-test init          # Deploy 9 Skills to your AI platform in one command
  ↓
/qa-test <requirement> # Start with natural language, AI executes 8 stages
  ↓
Requirement → Function → TestCase → Prepare → Execute → Review → Regression → Report

Core Philosophy

• Skills as Features — Each QA stage is a self-contained Skill file, read and executed by AI directly
• Methodology Built-in — Test case Skill embeds ISTQB methods: equivalence partitioning, boundary value analysis, error guessing
• HARD-GATE Confirmation — Every stage ends with a mandatory human confirmation gate
• Multi-Platform — Deploy the same Skills to OpenCode, Claude Code, Cursor

Architecture

┌─────────────────────────────────────────────────┐
│           AI Platform (OpenCode / Claude / Cursor)         │
│  Read Skill → Auto Execute → HARD-GATE Pause → Confirm     │
└──────────────────────────┬──────────────────────┘
                           │
           ┌───────────────┼───────────────┐
           ▼               ▼               ▼
    ┌──────────┐    ┌──────────┐    ┌──────────┐
    │ /qa-test │    │  Phase   │    │  Phase   │
    │   Entry  │───→│  Skill   │───→│  Skill   │
    └──────────┘    │   ×8     │    │   ×8     │
                    └──────────┘    └──────────┘
                           │
                           ▼
                  ┌─────────────────┐
                  │  .qa-test.yaml   │  ← State file (progress + resume)
                  │  Bash State Mgmt │  ← state / guard / handoff / archive
                  └─────────────────┘

8-Stage Workflow

| # | Skill | Trigger | Description | |---|-------|---------|-------------| | 1 | /qa-test-requirement | User input | Requirement analysis, Red Flags, follow-up questions | | 2 | /qa-test-function | Requirement confirmed | Feature breakdown, priority sorting | | 3 | /qa-test-testcase | Function confirmed | Test case generation (BVA + EP + Error Guessing) | | 4 | /qa-test-prepare | Test cases confirmed | Environment setup, sitemap, Playwright integration | | 5 | /qa-test-execute | Environment ready | Automated test execution, Playwright script generation | | 6 | /qa-test-review | Execution complete | Human review (OCR/Toast/Animation/Captcha) | | 7 | /qa-test-regression | Review complete | Failed case regression, script reuse | | 8 | /qa-test-report | Regression complete | Multi-format report generation, archiving |

Skill Structure

Each Skill uses a Decision Core + Reference Appendix two-layer structure:

skills/qa-test-testcase/
├── SKILL.md              # Decision Core: triggers, workflow, HARD-GATE
└── references/
    └── examples.md       # Reference Appendix: input/output examples

skills/methodology/       # Test methodologies (auto-loaded by Skills)
├── boundary-value.md     # Boundary Value Analysis
├── equivalence-partitioning.md  # Equivalence Partitioning
└── error-guessing.md     # Error Guessing + 12-category defect taxonomy

Quick Start

Prerequisites

• Node.js >= 18
• AI Platform: OpenCode / Claude Code / Cursor (any one)
• Git Bash (Windows users, scripts dependency)

Install

# Option 1: git clone + npm link (recommended, allows customizing Skills)
git clone <repo-url>
cd AI-QA-Agent-Platform
pnpm install --ignore-scripts
pnpm build
npm link

# Option 2: npm install from git
npm install -g github:<user>/qa-master

# Deploy Skills to your AI platform
qa-test init

qa-test init auto-detects your AI platform and copies 9 Skills to the right directory:

| Platform | Skills Location | |----------|----------------| | OpenCode | .opencode/skills/qa-test*/ | | Claude Code | .claude/skills/qa-test*/ | | Cursor | .cursor/skills/qa-test*/ |

Usage

qa-test init auto-detects your AI platform and copies the built-in skills/qa-test*/ to the right directory. Skills are distributed with the repo — no separate download needed.

# Start testing in your project (after Skills are deployed)
/qa-test Develop a user login feature with username/password validation and "Remember Me"

# Check progress
qa-test status

# No need to memorize commands — AI auto-advances through 8 stages, pausing at key checkpoints

State Management

All progress is tracked in .qa-test.yaml:

phase: testcase          # Current stage
testcase_artifact: artifacts/demo/testcase/testcase-20260705.md
confirmed_stages: [requirement, function]  # Confirmed stages

Supports resume — exit anytime, continue from the current stage on the next run.

Test Methodology

The platform ships with 3 ISTQB test design methodologies, auto-loaded during test case generation:

| Methodology | Key Contents | |-------------|-------------| | Boundary Value Analysis | 3-point method (min-1/min/max/max+1), covering 6 categories: numeric/string/collection/date/file/composite | | Equivalence Partitioning | Valid/invalid partitions, 5 partitioning methods (range/type/boolean/enum/time) | | Error Guessing | 12-category defect taxonomy (null/injection/concurrency/auth/network…), per-field inspection |

Multi-Platform Support

Same Skill definitions, deploy to different AI platforms:

# In an OpenCode project
cd my-opencode-project
qa-test init          # → .opencode/skills/qa-test*/

# In a Claude Code project
cd my-claude-project
qa-test init          # → .claude/skills/qa-test*/

# In a Cursor project
cd my-cursor-project
qa-test init          # → .cursor/skills/qa-test*/

Project Structure

├── skills/
│   ├── qa-test/                    # Entry Skill (phase detection + dispatch)
│   ├── qa-test-requirement/        # Requirement analysis
│   ├── qa-test-function/           # Feature breakdown
│   ├── qa-test-testcase/           # Test case generation (methodology built-in)
│   ├── qa-test-prepare/            # Environment setup (Playwright integration)
│   ├── qa-test-execute/            # Test execution
│   ├── qa-test-review/             # Human review
│   ├── qa-test-regression/         # Regression testing
│   ├── qa-test-report/             # Report generation
│   └── methodology/                # Test methodology references
├── scripts/
│   ├── qa-test-env.sh              # Script discovery
│   ├── qa-test-state.sh            # State management
│   ├── qa-test-guard.sh            # Phase guard
│   └── qa-test-archive.sh          # Test archiving
├── src/
│   ├── cli/                        # Installer (qa-test init / status)
│   ├── skills/loader.ts            # YAML frontmatter loader
│   ├── llm/                        # LLM integration
│   └── types.ts                    # Core type definitions
└── tests/
    └── generated/                  # Auto-generated Playwright scripts

Development

git clone <repo-url>
cd AI-QA-Agent-Platform
pnpm install
pnpm dev          # Dev mode (hot reload)
pnpm build        # Production build
pnpm test         # Run tests

License

MIT