feat: 使用 repository_dispatch 提高数据上限到 64KB

1. 修改 llm-bot-runner.yml:
   - 添加 repository_dispatch 事件支持 (types: [llm-task])
   - 重命名 job 为 prepare-context
   - 支持从 client_payload 读取数据
   - 当数据过大时从 issue 读取存储的数据

2. 修改 server.py:
   - 使用 repository_dispatch 触发工作流
   - 当数据超过 50KB 时，先创建 issue 存储完整数据
   - 在 payload 中传递 issue_number，工作流从 issue 读取数据
   - 添加 create_issue 辅助函数
   - 保留 workflow_dispatch 作为回退方案

Co-Authored-By: Claude (mimo-v2-flash) <noreply@anthropic.com>

Author: WhiteElephantIsNotARobot

.github/workflows/llm-bot-runner.yml

@@ -11,52 +11,81 @@ on:
         description: '任务的JSON上下文'
         required: true
         type: string
+  repository_dispatch:
+    types: [llm-task]
 
 jobs:
-  create-issue:
+  prepare-context:
     runs-on: ubuntu-latest
     permissions:
       issues: write
     outputs:
-      issue_number: ${{ steps.create-issue.outputs.issue_number }}
+      issue_number: ${{ steps.prepare-context.outputs.issue_number }}
+      task_content: ${{ steps.prepare-context.outputs.task_content }}
 
     steps:
-      - name: Create issue
-        id: create-issue
+      - name: Prepare context
+        id: prepare-context
         env:
           GH_TOKEN: ${{ github.token }}
-          TASK_CONTENT: ${{ inputs.task }}
-          CONTEXT_RAW: ${{ inputs.context }}
+          TASK_CONTENT: ${{ github.event.client_payload.task || inputs.task }}
+          CONTEXT_RAW: ${{ github.event.client_payload.context || inputs.context }}
+          ISSUE_NUMBER_FROM_PAYLOAD: ${{ github.event.client_payload.issue_number }}
           GITHUB_RUN_ID: ${{ github.run_id }}
           GITHUB_REPOSITORY: ${{ github.repository }}
+          EVENT_TYPE: ${{ github.event_name }}
         run: |
-          # 1. 格式化 Context
-          FORMATTED_CONTEXT=$(echo "$CONTEXT_RAW" | sed 's/^"//; s/"$//' | jq '.')
+          echo "Event type: $EVENT_TYPE"
 
-          # 2. 构建 Markdown 正文
-          printf "[工作流运行 #$GITHUB_RUN_ID](https://github.com/$GITHUB_REPOSITORY/actions/runs/$GITHUB_RUN_ID)\n\n## Task\n%s\n\n## Context\n\`\`\`json\n%s\n\`\`\`" \
+          # 1. 从 repository_dispatch 或 workflow_dispatch 获取数据
+          if [ "$EVENT_TYPE" = "repository_dispatch" ]; then
+            echo "Using repository_dispatch payload"
+
+            # 检查是否需要从 issue 读取（如果 payload 中的 context 为空或太大）
+            if [ -z "$CONTEXT_RAW" ] || [ "${#CONTEXT_RAW}" -gt 50000 ]; then
+              echo "Context from payload is empty or too large, reading from issue..."
+              # 从 payload 的 issue_number 字段获取存储数据的 issue
+              if [ -n "$ISSUE_NUMBER_FROM_PAYLOAD" ]; then
+                echo "Reading context from issue #$ISSUE_NUMBER_FROM_PAYLOAD"
+                # 从 issue body 读取数据
+                ISSUE_BODY=$(gh issue view "$ISSUE_NUMBER_FROM_PAYLOAD" --repo "$GITHUB_REPOSITORY" --json body --jq '.body')
+                # 提取 TASK 和 CONTEXT
+                TASK_CONTENT=$(echo "$ISSUE_BODY" | sed -n 's/.*<!-- TASK_START -->\(.*\)<!-- TASK_END -->.*/\1/p')
+                CONTEXT_RAW=$(echo "$ISSUE_BODY" | sed -n 's/.*<!-- CONTEXT_START -->\(.*\)<!-- CONTEXT_END -->.*/\1/p')
+              fi
+            fi
+          else
+            echo "Using workflow_dispatch inputs"
+          fi
+
+          # 2. 格式化 Context
+          FORMATTED_CONTEXT=$(echo "$CONTEXT_RAW" | sed 's/^"//; s/"$//' | jq '.' 2>/dev/null || echo "$CONTEXT_RAW")
+
+          # 3. 构建 Markdown 正文
+          printf "[工作流运行 #$GITHUB_RUN_ID](https://github.com/$GITHUB_REPOSITORY/actions/runs/$GITHUB_RUN_ID)\\n\\n## Task\\n%s\\n\\n## Context\\n\\`\\`\\`json\\n%s\\n\\`\\`\\`" \
             "$TASK_CONTENT" "$FORMATTED_CONTEXT" > body.md
 
-          # 3. 创建 Issue（硬编码标签 workflow）
+          # 4. 创建 Issue（硬编码标签 workflow）
           ISSUE_URL=$(gh issue create \
             --repo "$GITHUB_REPOSITORY" \
             --title "[工作流] Run $GITHUB_RUN_ID" \
             --body-file body.md \
             --label "workflow")
 
-          # 4. 提取 Issue 编号
+          # 5. 提取 Issue 编号
           ISSUE_NUMBER=${ISSUE_URL##*/}
           echo "issue_number=$ISSUE_NUMBER" >> "$GITHUB_OUTPUT"
+          echo "task_content=$TASK_CONTENT" >> "$GITHUB_OUTPUT"
 
           echo "成功创建 Issue: $ISSUE_NUMBER"
 
   run-llm-bot:
-    needs: create-issue
+    needs: prepare-context
     runs-on: ubuntu-latest
     env:
-      LLM_TASK: ${{ inputs.task }}
-      LLM_CONTEXT: ${{ inputs.context }}
-      LOG_ISSUE_NUMBER: ${{ needs.create-issue.outputs.issue_number }}
+      LLM_TASK: ${{ needs.prepare-context.outputs.task_content }}
+      LLM_CONTEXT: ${{ github.event.client_payload.context || inputs.context }}
+      LOG_ISSUE_NUMBER: ${{ needs.prepare-context.outputs.issue_number }}
       # GitHub身份令牌（4个名称都设置）- 使用agent的token
       GITHUB_TOKEN: ${{ secrets.WEINAR_API_KEY }}
       GH_TOKEN: ${{ secrets.WEINAR_API_KEY }}
@@ -213,7 +242,7 @@ jobs:
           echo ""
 
           # 构建新任务字符串，包含读取context.json的指示
-          ENHANCED_TASK=$(printf "%s\n\n> 任务上下文已保存到/app/context.json，共%s字符。阅读该文件以了解上下文！" "$LLM_TASK" "$CONTEXT_CHARS")
+          ENHANCED_TASK=$(printf "%s\\n\\n> 任务上下文已保存到/app/context.json，共%s字符。阅读该文件以了解上下文！" "$LLM_TASK" "$CONTEXT_CHARS")
           echo "增强任务: $ENHANCED_TASK"
 
           claude \

server.py

@@ -937,7 +937,7 @@ async def trigger_workflow(client: httpx.AsyncClient, ctx: TaskContext, task_tex
     # 检查上下文大小
     context_str = ctx.to_json_string()
     logger.info(f"Context size: {len(context_str)} chars")
-    
+
     # 调试信息
     if ctx.comments_history:
         logger.info(f"Comments history: {len(ctx.comments_history)} items")
@@ -951,35 +951,66 @@ async def trigger_workflow(client: httpx.AsyncClient, ctx: TaskContext, task_tex
             logger.info(f"  ReviewComment[{i}]: @{comment.get('user')} - {comment.get('path')}: {comment.get('body', '')[:50]}...")
     if ctx.is_truncated is not None:
         logger.info(f"Context was truncated: {ctx.is_truncated}")
-    
+
     # 检查是否有重复/空字段
     logger.info(f"diff_content present: {bool(ctx.diff_content)}")
     logger.info(f"clone_url: {ctx.clone_url}")
     logger.info(f"head_ref: {ctx.head_ref}, base_ref: {ctx.base_ref}")
     logger.info(f"head_repo: {ctx.head_repo}, base_repo: {ctx.base_repo}")
-    
+
     # 记录任务描述
     logger.info(f"LLM_TASK to send: '{task_text[:200]}{'...' if len(task_text) > 200 else ''}'")
-    
-    if len(context_str) > 60000:  # GitHub限制
-        logger.warning(f"Context too large ({len(context_str)} chars), truncating...")
-        # 简化上下文
-        ctx.diff_content = "[Diff truncated due to size limits]"
-        if ctx.comments_history and len(ctx.comments_history) > 10:
-            logger.info(f"Reducing comments history from {len(ctx.comments_history)} to 10 items")
-            ctx.comments_history = ctx.comments_history[-10:]  # 只保留最近10条
-        context_str = ctx.to_json_string()
+
+    # 检查是否需要将数据存储到 issue（因为 repository_dispatch 的 payload 也有 64KB 限制）
+    issue_number = None
+    use_repository_dispatch = True
+
+    # 如果数据太大，需要先创建 issue 存储数据
+    if len(context_str) > 50000 or len(task_text) > 10000:
+        logger.warning(f"Data too large (context: {len(context_str)}, task: {len(task_text)}), storing in issue...")
+
+        # 创建 issue 来存储数据
+        issue_body = f"""<!-- TASK_START -->{task_text}<!-- TASK_END -->
+<!-- CONTEXT_START -->{context_str}<!-- CONTEXT_END -->
+
+> 这个 issue 用于存储 workflow 的上下文数据，由 repository_dispatch 触发。
+> Run ID: {os.getenv('GITHUB_RUN_ID', 'unknown')}
+> Triggered by: {ctx.trigger_user}
+"""
+
+        issue_url = await create_issue(client, f"[数据存储] Context for node {node_id}", issue_body)
+        if issue_url:
+            issue_number = issue_url.split('/')[-1]
+            logger.info(f"Created issue #{issue_number} to store context data")
+            # 清空数据，因为将从 issue 读取
+            context_str = ""
+            task_text = ""
+        else:
+            logger.error("Failed to create issue for large data, falling back to truncation")
+            # 回退到截断方案
+            use_repository_dispatch = False
 
     url = f"{REST_API}/repos/{CONTROL_REPO}/actions/workflows/llm-bot-runner.yml/dispatches"
     headers = {"Authorization": f"token {GQL_TOKEN}", "Accept": "application/vnd.github.v3+json"}
 
+    # 构建 payload
     payload = {
         "ref": "main",
-        "inputs": {
+        "client_payload": {}
+    }
+
+    if use_repository_dispatch:
+        # 使用 repository_dispatch
+        payload["client_payload"]["task"] = task_text[:2000] if task_text else ""
+        payload["client_payload"]["context"] = context_str if context_str else ""
+        if issue_number:
+            payload["client_payload"]["issue_number"] = issue_number
+    else:
+        # 回退到 workflow_dispatch（数据已截断）
+        payload["inputs"] = {
             "task": task_text[:2000],
             "context": context_str
         }
-    }
 
     try:
         r = await client.post(url, headers=headers, json=payload)
@@ -1012,6 +1043,32 @@ async def trigger_workflow(client: httpx.AsyncClient, ctx: TaskContext, task_tex
         logger.error(f"Exception during workflow dispatch: {e}")
         return False
 
+
+async def create_issue(client: httpx.AsyncClient, title: str, body: str) -> Optional[str]:
+    """
+    创建 GitHub Issue 并返回 URL
+    """
+    url = f"{REST_API}/repos/{CONTROL_REPO}/issues"
+    headers = {"Authorization": f"token {GQL_TOKEN}", "Accept": "application/vnd.github.v3+json"}
+
+    payload = {
+        "title": title,
+        "body": body,
+        "labels": ["context-data"]
+    }
+
+    try:
+        r = await client.post(url, headers=headers, json=payload)
+        if r.status_code == 201:
+            data = r.json()
+            return data.get("html_url")
+        else:
+            logger.error(f"Failed to create issue: {r.status_code} - {r.text}")
+            return None
+    except Exception as e:
+        logger.error(f"Exception creating issue: {e}")
+        return None
+
 # --- 轮询逻辑 ---
 async def poll_loop():
     """