← Back to Skills Marketplace
loveshaozhe

qa-req2testcase-generator

by loveshaozhe · GitHub ↗ · v4.12.6 · MIT-0
cross-platform ⚠ suspicious
29
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install qa-req2testcase-generator
Description
AI驱动的需求→测试用例生成能力。V4.12.6架构:总控路由+逐条P6生成+渐进式披露+P6质量引导+反脚本防护+批量修复
README (SKILL.md)

qa-req2testcase-generator V4.12.6

版本:4.12.6 | 架构:总控路由+逐条P6生成+渐进式披露+P6质量引导+反脚本防护+批量修复 V4.12.6关键变更: ⚖️ p6_merge比率自动调平+🔧 p7_batch_fix批量修复+📤 评审推送默认开启 | V4.12.5: 🛡️ 反脚本三段式拦截

⚠️ 元规则（最高优先级，违反即终止）

Agent必须100%执行skill规定的流程，不得自行判断"优化"或"改进"。

  1. 执行优先：按skill规定流程执行每一步，不得跳过、绕道
  2. 规则即硬约束：「禁止」「必须」「🔴」是代码级硬约束，违反即报错
  3. 发现问题先执行后反馈：完成当前流程后反馈，不在执行中自行修改
  4. 禁止伪造结果：orchestrator返回error时必须修复重试，不得伪造gate pass
  5. 禁止自我决策：不允许Agent自行判断"规则不合理/太慢/可优化"并绕过
  6. 禁止抛选择题：任何步骤失败后自动修复重试，禁止停下来向用户要选择
  7. ⛔ 禁止脚本批量生成（V4.12.5）：禁止编写 Python/Shell/JavaScript 脚本循环调用 p6_generate_one。这不是效率，是偷懒。脚本批量生成的内容必然空洞，质量门禁会拦截，拦截后再重来反而更慢。必须由 LLM 逐条阅读 prompt 后手写生成每个 TP 的用例。orchestrator 会检测调用频率，违规将被拒绝。

⛔ 入口强制检查：需求载荷是否存在（最高优先级）

此检查在技能触发后第一个动作执行。不通过→立即终止。

✅ 允许继续：用户消息含需求正文 / 含需求附件(.docx/.txt/.pdf) / 明确引用近期需求 ❌ 必须停止："我等会发"、"分析这个需求"(无附件)、仅提及技能名、模糊引用

不通过回复：📋 请先发送需求正文或需求文档，收到后立即开始分析。 通过回复：✅ 需求载荷已收到 | 格式:X | 大小:Y | 📋 即将进入初始化...

🔴 运行协议

核心原则:orchestrator.py控制流程,Agent只负责执行prompt返回JSON。

6段确认模式:每次「继续」只前进一个段落（绝对禁止连段执行）:

  • 用户回复1次「继续」→ Agent只执行下一个段落，执行完立即⏸️停止
  • ⛔ 禁止: 用户说1次「继续」Agent连跑多段
  • ⛔ 禁止: 看到下一段gate已存在就"继续跑完"
  • 每段结束后检查: 是否已输出⏸️？是否已等待「继续」？

📌 __must_emit__ 字段：p2_code_generate / step7_export的stdout包含此字段。Agent必须将MEDIA行复制到回复，同时用 exec cat 展示文件内容。

🔴 段落边界规则（最高优先级，违反视为流程无效）

规则 说明
⛔ 禁止连段 每段结束→⏸️停止→等「继续」→才读下一段规则文件
⛔ 禁止预读 执行段落N时，禁止读取 rules/paragraph_N+1.md
⛔ 禁止跨段 即使下一段gate已存在，也必须等用户「继续」
🔴 停止锚点 每段规则文件末尾有终止锚点，Agent读到必须停止

🔴 操作约束矩阵

Step Agent角色 唯一正确命令
P0 执行者 prep_prompt → 生成JSON → step_run
P1 执行者(分批) 骨架→循环feature→p1_code_merge
P2 观察者 python3 "$ORCH" --action p2_code_generate
P3 执行者 prep_prompt → 生成JSON → step_run
P4 执行者 prep_prompt → 生成JSON → step_run
P5 观察者 python3 "$ORCH" --action p5_code_merge
P6 LLM生成者(⛔禁子Agent) p6_tp_list→逐条循环(p6_generate_one→完整阅读prompt生成JSON→p6_generate_one --save)→p6_merge\x3Cbr>🔴 V4.11.0逐条生成：一次一个测试点，prompt极简(~400B)。Agent完整阅读prompt后生成JSON，字段由代码自动补全
P7 观察者 python3 "$ORCH" --action p7_code_check

Agent只做3类事: ①exec orchestrator命令 ②read prompt并生成JSON write到文件 ③read图片并描述

🔴🔴🔴 V3.5.2 绝对禁止行为（代码层硬控）

  1. 禁止直接写gate文件 — gates/*.pass.json 只能由orchestrator创建
  2. 禁止直接写output文件 — p{N}_output.json 只能通过step_run等action写入
  3. 禁止直接修改state文件 — orchestrator_state.json 内部修改
  4. 禁止import orchestrator — 不允许任何形式导入
  5. 禁止跳过orchestrator — 所有步骤必须通过 python3 "$ORCH" --action XXX
  6. 禁止伪造执行结果 — error时必须修复重试，不得手动替代
  7. Onboarding必须逐步交互 — 3步逐步展示，禁止合并

违反以上任何规则，step7_export会审计拒绝。

📋 6段执行流程

段落 内容 规则文件 前置Gate 确认点
1 init + onboarding read rules/paragraph_1.md — ⏸️
2 step0 + 图片理解 read rules/paragraph_2.md P0 gate ⏸️
3 P0+P1 + 自动P2 read rules/paragraph_3.md step0 gate ⏸️
4 P3+P4 + 自动P5 read rules/paragraph_4.md P2 gate ⏸️
5 P6 用例生成 read rules/paragraph_5.md P5 gate ⏸️
6 P7 + Excel导出 read rules/paragraph_6.md P6 gate 🏁

🔴 执行流程（每段通用，严格执行）:

  1. 收到用户「继续」→ 查上表找到下一段N
  2. read rules/paragraph_N.md → 完整阅读该段规则
  3. 严格按规则文件中的指令逐步执行
  4. 执行到规则文件末尾的终止锚点 → ⏸️停止
  5. 输出「段落N完成，请回复「继续」」
  6. 🔴 绝对禁止：读完paragraph_N.md后继续读取paragraph_N+1.md

🔄 断点续跑

exec: python3 "$ORCH" --action status

→ 查看已完成的gate pass，从下一个未执行的步骤开始。 → 如果当前段落已有部分gate pass → 阅读该段规则文件，跳过已完成步骤。

❌ 错误处理

错误类型 处理
gate_blocked 检查缺失的前置步骤，从该步骤重新执行
guard_failed 检查truncation，修复JSON后重试
quality_rejected 按issues和fix_example修复，最多重试2次
timeout 检查文件是否已生成新内容，有则继续，无则重试

📁 文件结构

skill_v4/
├── SKILL.md                     ← 本文件（总控路由）
├── rules/
│   ├── paragraph_1.md           ← 段落1规则
│   ├── paragraph_2.md           ← 段落2规则
│   ├── paragraph_3.md           ← 段落3规则
│   ├── paragraph_4.md           ← 段落4规则
│   ├── paragraph_5.md           ← 段落5规则
│   └── paragraph_6.md           ← 段落6规则
├── prompts/                     ← LLM prompt模板
├── tools/                       ← orchestrator.py等工具
├── config/                      ← 配置
└── references/                  ← 参考文档

触发条件

「ai用例生成」「ai需求分析」「req2testcase」「生成测试用例」「分析需求」「拆解功能点」「输出测试点」「需求评审」「PRD转测试用例」

已知限制

  • 仅支持中文需求文档（.docx/.txt/粘贴文本）
  • 不支持视频/音频需求输入
  • PX图片理解依赖腾讯云API（未配置则降级为caption_only）
  • LOW模型（MiniMax等）V4.10.0起走窄聚焦模式，用例质量可达可用水平
Usage Guidance
Install only if you are comfortable with this skill searching local folders for requirement files, installing Python packages, storing API credentials, and sending requirement-derived summaries or test cases to the configured remote HTTP review service. Prefer disabling cloud push/sync, removing the hardcoded API key, requiring explicit file paths, and using a controlled environment before use.
Capability Tags
cryptorequires-walletrequires-paid-servicerequires-sensitive-credentials
Capability Assessment
⚠ Purpose & Capability
The core requirement-to-testcase workflow is coherent, but artifacts expand it into OCR/image API use, online review push, cloud knowledge sync, broad document discovery, and persistent learning. Those capabilities handle sensitive requirement and test data and are broader than the headline generator purpose.
⚠ Instruction Scope
The skill gives agents strict execution commands, broad filesystem searches, and non-interactive retry/continue behavior. It also contains conflicting disclosure: user-facing text says the API password is not saved, while code caches it and writes it into cloud.json.
⚠ Install Mechanism
The package includes runtime dependency installation instructions and an import-time pip install in tools/image_extract.py. That mutates the host environment during normal use rather than relying on declared prerequisites.
⚠ Credentials
Searching common user folders and copying recent documents is overbroad for processing an explicitly supplied requirement file. External HTTP endpoints and image-analysis services are enabled in configuration, including a hardcoded API key.
⚠ Persistence & Privilege
The skill persists state in user_knowledge, .image_api_key, cloud.json, cloud-synced knowledge files, push markers, and retry queues. Automatic review push can send generated cases and summaries to a remote service when an API key is present.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install qa-req2testcase-generator
  3. After installation, invoke the skill by name or use /qa-req2testcase-generator
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v4.12.6
V4.12.6: p6_merge比率自动调平+p7_batch_fix批量修复+评审推送默认开启+Excel导出修复+评审工具500修复
Metadata
Slug qa-req2testcase-generator
Version 4.12.6
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is qa-req2testcase-generator?

AI驱动的需求→测试用例生成能力。V4.12.6架构:总控路由+逐条P6生成+渐进式披露+P6质量引导+反脚本防护+批量修复. It is an AI Agent Skill for Claude Code / OpenClaw, with 29 downloads so far.

How do I install qa-req2testcase-generator?

Run "/install qa-req2testcase-generator" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is qa-req2testcase-generator free?

Yes, qa-req2testcase-generator is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does qa-req2testcase-generator support?

qa-req2testcase-generator is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created qa-req2testcase-generator?

It is built and maintained by loveshaozhe (@loveshaozhe); the current version is v4.12.6.

More Skills
self-improving agent
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
⬇ 463283 · by pskoett
Skill Vetter
Security-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
⬇ 259410 · by spclaudehome
Polymarket
Query Polymarket prediction markets. Check odds, find trending markets, search events, track price movements.
⬇ 232503 · by joelchance
Self-Improving + Proactive Agent
Self-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
⬇ 200423 · by ivangdavila
Github
Interact with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.
⬇ 191115 · by steipete
ontology
Typed knowledge graph for structured agent memory and composable skills. Use when creating/querying entities (Person, Project, Task, Event, Document), linkin...
⬇ 190156 · by oswalpalash

💬 Comments