gherkin-generator-skill

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Skill: Gherkin Generator (BDD)

技能：Gherkin生成器（BDD）

Converts functional requirements into valid, high-quality Gherkin scenarios through an iterative cycle with two specialised agents.

通过双专业Agent的迭代循环，将功能需求转换为合规、高质量的Gherkin场景。

System Architecture

系统架构

Functional requirements
        │
        ▼
┌─────────────────────┐
│  gherkin-generator  │  ← Agent 1: generates/modifies the .feature file
│       -agent        │
└────────┬────────────┘
         │ Gherkin draft
         ▼
┌─────────────────────┐
│  gherkin-reviewer   │  ← Agent 2: validates and proposes improvements
│       -agent        │
└────────┬────────────┘
         │
    Approved?
    NO ──────────────────► returns to Agent 1 with feedback
    YES ─────────────────► delivers the final .feature file

Functional requirements
        │
        ▼
┌─────────────────────┐
│  gherkin-generator  │  ← Agent 1: 生成/修改.feature文件
│       -agent        │
└────────┬────────────┘
         │ Gherkin draft
         ▼
┌─────────────────────┐
│  gherkin-reviewer   │  ← Agent 2: 验证并提出改进建议
│       -agent        │
└────────┬────────────┘
         │
    Approved?
    NO ──────────────────► 携带反馈返回给Agent 1
    YES ─────────────────► 交付最终.feature文件

Step-by-step workflow

分步工作流程

Step 1 — Prepare the requirements

步骤1 — 准备需求

Before invoking the agents, extract from the user's input:

Module / feature name (for the Feature name)
List of functional requirements (numbered or in prose)
Business context (actors, systems involved)
Example data if available

If anything is ambiguous, ask the user before starting.

调用Agent前，从用户输入中提取以下信息：

模块/功能名称（用于Feature名称）
功能需求列表（编号或散文形式）
业务上下文（涉及的角色、系统）
示例数据（如果有）

如果存在不明确的内容，在开始前先询问用户。

Step 2 — Invoke the Generator Agent

步骤2 — 调用生成Agent

Read the full instructions in

agents/generator-agent.md

Call the agent with:

INPUT:
  requirements:      <list of requirements>
  context:           <business context>
  previous_feedback: <empty on first iteration>
  previous_gherkin:  <empty on first iteration>
  iteration:         <current iteration number>

The agent returns a Gherkin block in

.feature

format.

阅读

agents/generator-agent.md

中的完整说明。

调用Agent时传入：

INPUT:
  requirements:      <需求列表>
  context:           <业务上下文>
  previous_feedback: <首次迭代时为空>
  previous_gherkin:  <首次迭代时为空>
  iteration:         <当前迭代次数>

Agent会返回.feature格式的Gherkin代码块。

Step 3 — Invoke the Reviewer Agent

步骤3 — 调用评审Agent

Read the full instructions in

agents/reviewer-agent.md

Call the agent with:

INPUT:
  gherkin:               <output from the generator agent>
  original_requirements: <user's original requirements>
  iteration:             <current iteration number>

The agent returns an object with:

```
approved: true/false
```
```
score: 0-100
```
```
issues: [list of problems]
```

suggestions: [list of concrete improvements]

annotated_gherkin: <the gherkin with inline comments>

阅读

agents/reviewer-agent.md

中的完整说明。

调用Agent时传入：

INPUT:
  gherkin:               <生成Agent的输出内容>
  original_requirements: <用户的原始需求>
  iteration:             <当前迭代次数>

Agent会返回一个包含以下字段的对象：

```
approved: true/false
```
```
score: 0-100
```
```
issues: [问题列表]
```
```
suggestions: [具体改进建议列表]
```

annotated_gherkin: <带有内联注释的Gherkin内容>

Step 4 — Improvement loop

步骤4 — 改进循环

iteration = 1
maximum   = 5   ← prevents infinite loops

WHILE iteration <= maximum:
  gherkin  = generator_agent(requirements, feedback, previous_gherkin)
  review   = reviewer_agent(gherkin, requirements)

  IF review.approved == true OR review.score >= 85:
    BREAK

  feedback         = review.suggestions + review.issues
  previous_gherkin = gherkin
  iteration       += 1

IF iteration > maximum:
  deliver the best gherkin obtained with a warning note

iteration = 1
maximum   = 5   ← 防止无限循环

WHILE iteration <= maximum:
  gherkin  = generator_agent(requirements, feedback, previous_gherkin)
  review   = reviewer_agent(gherkin, requirements)

  IF review.approved == true OR review.score >= 85:
    BREAK

  feedback         = review.suggestions + review.issues
  previous_gherkin = gherkin
  iteration       += 1

IF iteration > maximum:
  交付已得到的最优Gherkin内容，并附带警告说明

Step 5 — Delivery

步骤5 — 交付

Present to the user:

The final
```
.feature
```
file with syntax highlighting
The reviewer's final score
A summary of iterations performed
(Optional) Save the file if the user requests it

向用户展示以下内容：

带语法高亮的最终.feature文件
评审Agent的最终评分
执行的迭代次数总结 4.（可选）如果用户要求，保存该文件

Approval criteria (summary)

评审通过标准（摘要）

The Reviewer Agent approves when all of the following are met:

Criterion	Minimum requirement
Requirements coverage	100% of functional requirements covered
Gherkin syntax	No errors (Feature, Scenario, Given/When/Then)
Scenario independence	No order dependency between scenarios
Step clarity	No technical ambiguities
Example data	Scenario Outline + Examples tables when variants exist
Language	Consistent throughout the file

当所有以下条件满足时，评审Agent会通过：

评审标准	最低要求
需求覆盖率	100%覆盖所有功能需求
Gherkin语法	无语法错误（Feature、Scenario、Given/When/Then）
场景独立性	场景之间无顺序依赖
步骤清晰度	无技术歧义
示例数据	存在变体时使用Scenario Outline + Examples表格
语言一致性	文件内语言保持一致

Implementation notes

实现说明

Maximum iterations: 5 (configurable by the user)
Language: use the same language as the input requirements
Granularity: one
```
.feature
```
file per module/feature
Tags: add
```
@smoke
```
,
```
@regression
```
,
```
@happy-path
```
,
```
@edge-case
```
as appropriate
Background: use only when ≥3 scenarios share the same Given steps

For implementation details of each agent, read:

```
agents/generator-agent.md
```
```
agents/reviewer-agent.md
```
```
references/gherkin-best-practices.md
```

最大迭代次数：5次（用户可配置）
语言：使用与输入需求相同的语言
粒度：每个模块/功能对应一个.feature文件
标签：根据情况添加
```
@smoke
```
、
```
@regression
```
、
```
@happy-path
```
、
```
@edge-case
```
Background：仅当≥3个场景共享相同的Given步骤时使用

关于每个Agent的实现细节，请阅读：

```
agents/generator-agent.md
```
```
agents/reviewer-agent.md
```
```
references/gherkin-best-practices.md
```