🤖 Test-Agent

AI Testing Agent Framework · Open-Source · Multi-LLM · 5-second setup

English | 简体中文

---

⚡ 30-second demo

git clone https://github.com/Wool-xing/Test-Agent.git
cd Test-Agent && pip install -e .

tagent demo            # 0 API key · 0 config · stub LLM · 30s end-to-end

Outputs: test cases (Excel + xmind + markmap + opml) + Word report + decision logs, all under workspace/.

Ready to run on your project?

tagent init --preset 国内-web    # or: minimal / saas-web / mobile-android / security-pentest
# → produces .env + tagent.yml + STARTUP.md (5-step onboarding guide)

8640 config combinations from a single matrix.yaml — change a line in YAML, the wizard picks it up. See 04-配置文件/templates/INDEX.md.

---

What is Test-Agent?

Test-Agent turns any software, EXE, APK, Docker image, or API into a fully tested project — autonomous from requirement parsing to PoC-validated bug reports. Built for QA teams, security researchers, automotive testers, and anyone who wants to use AI testing while learning the theory behind it.

• 16 expert agents — functional · security · mobile · desktop · AI model · automotive · pentest …
• 32+ reusable skills — TDD · E2E · regression · pentest · car-CAN-bus · eval-harness · …
• 49 production utils — pytest · Playwright · JMeter · Appium · Burp · Allure · OpenCV · …
• Multi-LLM — Claude / OpenAI / Gemini / Qwen / DeepSeek / Ollama(local,no vendor lock-in)
• 6 BugTracker adapters — Zentao · Jira · GitHub Issues · GitLab Issues · Linear · Webhook(主宪章 §37)
• 6 notify channels — WeChat Work · Lark/Feishu · DingTalk · Slack · Email · MS Teams(主宪章 §36)
• MCP-native — 6-server suite + 4-gate marketplace
• 4-layer self-test — L1 lint · L2 mock CI · L3 real-LLM pre-tag · L4 weekly cron(主宪章 §33)

🚀 Install for production

curl -fsSL https://raw.githubusercontent.com/Wool-xing/Test-Agent/main/install.sh | bash -s -- /path/to/your-test-project

Then tagent init to scaffold .env/tagent.yml/STARTUP.md — no more 30 mins of hand-editing.

🎯 5 Key Capabilities

1. All-platform — Web / API / Android / iOS / WeChat-miniprogram / Windows EXE / macOS / Linux / Electron / game / IoT / audio-video / AI/LLM / blockchain / 车载
2. All-protocol — HTTP(S) / gRPC / WebSocket / TCP / UDP / GraphQL / SOAP / MQTT / SSH / serial / Kafka / RabbitMQ / Modbus / CAN-bus / SOME-IP / DoIP / UDS
3. Multi-LLM no lock-in — switch with tagent model between Claude / OpenAI / Gemini / Qwen / DeepSeek / Ollama
4. Learn while using — --mode learn outputs every step with theory references (22 KB cards across 13 domains: tools / coding / foundation / strategy / methods / protocols / platforms / gates / security / AI testing / compliance / process / build-your-own)
5. Safe-by-default — sandboxed exec / PII scrub / runtime prompt-injection scan / 4-gate marketplace verify / decisions audit trail

📊 Coverage

• Product types: Web · API · Mobile · Desktop · IoT · AI · Blockchain · Vehicle · Embedded · Serverless
• Test types: functional / performance / security / compatibility / weak-network / stability / reliability / accessibility / contract / visual / i18n / observability / chaos / mutation / AI-specific (hallucination / prompt-injection / drift / fairness) / compliance
• Test design methods: equivalence-partitioning · boundary-value · decision-table · state-transition · pairwise · orthogonal · exploratory SBTM · risk-based · TDD · BDD · ATDD
• Quality gates: smoke → regression → performance_ci_quick → performance_full → release (5-layer)

Total ≈ 95% coverage — remaining 5% (DO-178C avionics / HIPAA medical / IEC 61508 industrial) added by your domain experts.

🏛️ Charter-Driven

Test-Agent ships with a 31-section charter (CHARTER.md-equivalent) covering:

• §10–§12 · Soul (3 axioms + 5 inscriptions + 16 key terms)
• §13–§17 · Architecture (experts / skills / installs / darwin self-evolution / AgentChat / MCP)
• §18–§21 · Methodology (9-cluster map / test pyramid 2024 / 18 closed-loop rules / 9-industry adapter / 50+ test types / 4 depth levels)
• §22 · Hermes-inspired (scheduler / subagent / learning-loop / 7 backends / 8 platforms)
• §23 · Teaching layer (KB 13 categories + anti-hallucination 3 layers + bilingual)
• §24 · GBrain-inspired (KB self-wiring graph + eval replay + PII scrub)
• §25–§26 · Pentest & Automotive verticals
• §27 · Karpathy 4 principles (think-before / simplicity-first / surgical / goal-driven)
• §28 · ECC test hardening (tdd-workflow / verification-loop / e2e / eval-harness / security-review)
• §29 · Essence watcher (auto-track upstream OSS for delta extraction)
• §30 · Marketplace 4-lane (4-gate security)
• §31 · Build-your-own-X learning layer

📂 Project Structure

Test-Agent/
├── 00-项目导航.md           ← 5-dimension category guide
├── 01-快速开始/             ← user manual / deploy / config / deliverables
├── 02-专家定义/             ← 16 expert agents
├── 03-技能定义/             ← 34 skills (incl. darwin-skill / karpathy-guidelines upstream)
├── 04-配置文件/             ← conftest / pytest.ini / .env / .mcp.json
├── 05-代码示例/             ← 49 production utils
├── 06-CICD集成/             ← GitHub Actions + Jenkins
├── runtime/                ← V1.x runtime layer (router / orchestrator / MCP / web / scheduler / subagent / learning_loop / backends / gateway / tutor / essence_watcher / marketplace)
├── docs/theory/            ← 22 teaching KB cards across 13 categories
├── profiles/compliance/    ← 10 industry compliance YAML profiles
├── marketplace/            ← Community skills / agents / mcp / hooks (4 lanes, 4-gate verify)
├── install.sh              ← one-line deploy
├── README.md               ← This file
├── FULL_GUIDE.md           ← Full engineering guide
├── CHANGELOG.md            ← Version log
└── LICENSE / SECURITY.md / CONTRIBUTING.md / CODE_OF_CONDUCT.md

📚 Documentation

Audience	Read
First-time user	Quick start → Deploy
QA engineer	User manual → Skill catalog
Architect / SRE	Architecture deep-dive → Runtime
Security researcher	Pentest expert → pentest-coordinator
Automotive tester	Automotive expert → ASIL workflow
Contributor	CONTRIBUTING.md → Marketplace

🛠️ Tech Stack

pytest 8.3 · Playwright 1.59 · Appium 5.3 · pywinauto · JMeter 5.6 · Allure · Airtest · OpenCV · Faker · SQLAlchemy 2.0 · MCP 1.0 · LiteLLM · Prefect · FastAPI · React 18 · Tailwind · Postgres+pgvector · MinIO · OpenTelemetry · Loguru · Docker Compose · GitHub Actions / Jenkins

🤝 Contributing

See CONTRIBUTING.md for the full workflow (sync rules + RACI matrix + 6-layer dependency policy + Karpathy 4 principles).

Community marketplace contributions (marketplace/) go through 4 safety gates: signature → injection scan → docker sandbox → darwin-skill scoring.

📜 License

MIT License — see LICENSE.

Upstream components retain their own licenses; see NOTICE.md for attributions.

🙏 Inspirations (essence absorbed)

• hermes-agent — closed learning loop + 7 backends + multi-platform gateway
• gbrain — self-wiring KB graph + eval replay + safe-by-default
• andrej-karpathy-skills — 4 LLM-coding principles
• everything-claude-code — TDD / verification / harness-first
• pentagi + shannon — pentest agent black-box + white-box
• build-your-own-x — deep-dive learning path

---

Made for testers · Built with testers · Tested by testers