用 OpenRouter GPT 自动评审 Jira 工单质量并添加辅导评论

Jira 触发 → AI 质量评分 → 低分自动评论

给谁

使用 Jira Cloud 的团队负责人或项目经理

你会得到

新 Story/Bug/Task 创建时自动评估质量，低分 ticket 收到辅导式改进建议

你需要

n8n · jira

步骤

1. 配置触发器与阈值

   在 n8n 中导入模板后，配置 Jira Trigger 节点：选择 Jira Software Cloud 凭证，设置事件过滤为 issue_created 且 issuetype in (Story, Bug, Task)。然后配置 Set Score Threshold 节点，将 SCORE_THRESHOLD 设为所需阈值（默认 7）。

2. 配置 AI 质量评估

   连接 OpenAI Quality Evaluator（OpenRouter Chat Model）到 Quality Analysis Agent。在 Quality Analysis Agent 中确认提示词：user message 传入 ticket 类型、标题、描述和报告人；system message 定义 5 项评分标准（标题清晰度、描述完整性、验收标准、重现步骤/用户场景、范围与依赖），每项 1-3 分，输出指定 JSON 结构。

   运行提示词 · 来源原文。Quality Analysis Agent 节点的 Text / User Message 字段

   复制提示词

   =Evaluate this Jira ticket:
   
   <ticket>
   Type: {{ $('When JIRA Ticket Created').first().json.issue.fields.issuetype.name }}
   Title: {{ $('When JIRA Ticket Created').first().json.issue.fields.summary }}
   Description: {{ $('When JIRA Ticket Created').first().json.issue.fields.description ?? "No description provided." }}
   Reporter: {{ $('When JIRA Ticket Created').first().json.issue.fields.reporter.displayName }}
   </ticket>
   
   Return the evaluation JSON.

   运行提示词 · 来源原文。Quality Analysis Agent 节点的 System Message 字段

   复制提示词

   You are a Jira ticket quality reviewer. Your role is to evaluate whether 
   a ticket contains enough information for a developer to start working on it 
   without asking clarifying questions.
   
   Evaluate the ticket on these 5 criteria:
   
   1. **Title clarity** — Is the title specific and action-oriented? (not "Fix bug" or "Update page")
   2. **Description completeness** — Does it explain the context, the problem, and the expected behavior?
   3. **Acceptance criteria** — Are there clear, testable conditions for "done"?
   4. **Reproduction steps** — For bugs: are steps to reproduce present? For stories: is there a user scenario?
   5. **Scope & dependencies** — Are unknowns, blockers, or related tickets mentioned?
   
   For each criterion, assign a score from 1 to 3:
   - 1 = Missing or too vague to be useful
   - 2 = Present but incomplete
   - 3 = Clear and sufficient
   
   Output a JSON object with this exact structure:
   {
     "scores": {
       "title_clarity": 1-3,
       "description_completeness": 1-3,
       "acceptance_criteria": 1-3,
       "reproduction_steps": 1-3,
       "scope_dependencies": 1-3
     },
     "total": 5-15,
     "normalized_score": 1-10,
     "missing": ["list of the criteria scored 1"],
     "comment": "2-3 sentence coaching comment addressed directly to the ticket author. Friendly tone. Name exactly what is missing and why it matters. Do not mention the score."
   }
   
   Return only the JSON object. No preamble, no markdown fences.

3. 配置分数解析

   在 Parse Normalized Score 节点（Code 节点）中保留默认 JS 代码：从 AI 输出中提取 JSON，去除 markdown 代码块标记，解析为对象并返回。若解析失败则抛出错误。

   运行提示词 · 来源原文。Parse Normalized Score 节点的 JS Code 字段

   复制提示词

   const response = $input.first().json.output ?? $input.first().json.text ?? "";
   
   const cleaned = response
     .replace(/^```json\s*/i, "")
     .replace(/^```\s*/i, "")
     .replace(/\s*```$/i, "")
     .trim();
   
   let scores;
   
   try {
     scores = JSON.parse(cleaned);
   } catch (error) {
     throw new Error(`Invalid JSON returned by Quality Checker Agent.\n${error.message}`);
   }
   
   return scores;

4. 配置条件分支

   在 Check Score Threshold 节点（If 节点）中设置条件：normalized_score < SCORE_THRESHOLD。确保左侧值取自 Parse Normalized Score 的输出，右侧值取自 Set Score Threshold。

5. 配置评论动作

   在 Add Comment to JIRA Ticket 节点中配置 Jira 凭证，确认评论模板包含辅导建议、质量分数和缺失项列表。若分数达标，工作流走向 No Action Needed 节点结束。

   运行提示词 · 来源原文。Add Comment to JIRA Ticket 节点的 Comment 字段

   复制提示词

   ={{ $json.comment }}
   
   **Quality score: {{ $json.normalized_score }}/10**
   Missing: {{ $json.missing.join(', ') }}
   
   _This comment was generated automatically to help keep our backlog clean._

你是一个 n8n 工作流搭建助手。用户想要复现一个 Jira 工单质量自动评审工作流。请协助用户：1）确认已具备 Jira Cloud 和 OpenRouter 凭证；2）从模板 https://n8n.io/workflows/16693/ 导入工作流；3）按步骤检查节点配置，特别是 Quality Analysis Agent 的 system message 和 OpenRouter 模型选择；4）测试时先创建一个低质量 Jira ticket 验证评论是否触发。注意：默认模型为 openai/gpt-oss-20b:free，可根据需要更换。