@fitlab-ai/agent-infra 0.5.8 → 0.5.9

package/templates/.agents/rules/issue-sync.github.en.md

@@ -115,7 +115,7 @@ If `gh` fails, skip and continue. Do not fail the skill.
 
 When a skill creates or imports an Issue, automatically add the current executor as assignee:
 
-- `create-issue`: use `--assignee @me` in `gh issue create` and include `-R "$upstream_repo"`
+- When the `create-task` platform rule triggers Issue creation: use `--assignee @me` in `gh issue create` and include `-R "$upstream_repo"`
 - `import-issue`: run `gh issue edit {issue-number} -R "$upstream_repo" --add-assignee @me 2>/dev/null || true` after import
 
 `@me` is resolved by `gh` CLI to the authenticated user. The operation is idempotent. If the command fails, skip it directly and do not provide a fallback path.

package/templates/.agents/rules/issue-sync.github.zh-CN.md

@@ -115,7 +115,7 @@ fi
 
 当技能创建或导入 Issue 时，自动将当前执行者添加为 assignee：
 
-- `create-issue`：在 `gh issue create` 命令中使用 `--assignee @me` 参数，并附带 `-R "$upstream_repo"`
+- `create-task` 的平台规则触发 Issue 创建时：在 `gh issue create` 命令中使用 `--assignee @me` 参数，并附带 `-R "$upstream_repo"`
 - `import-issue`：导入后执行 `gh issue edit {issue-number} -R "$upstream_repo" --add-assignee @me 2>/dev/null || true`
 
 `@me` 由 `gh` CLI 自动解析为当前认证用户。此操作是幂等的（重复添加不会报错）。如果命令失败（如权限不足），直接跳过，不做任何替代。

package/templates/.agents/rules/milestone-inference.github.en.md

@@ -1,6 +1,6 @@
 # Milestone Inference Rules
 
-Read this file before `create-issue`, `implement-task`, or `create-pr` handles a milestone.
+Read this file before the `create-task` platform rule, `implement-task`, or `create-pr` handles a milestone.
 
 ## General Principles
 
@@ -21,7 +21,7 @@ git branch -r | grep -v 'HEAD' | grep -E 'origin/[0-9]+\.[0-9]+\.x$'
 - Any output: multi-version branch mode
 - No output: trunk mode
 
-## Phase 1: `create-issue`
+## Phase 1: `create-task` (when the platform rule creates an Issue)
 
 Goal: choose a coarse-grained release line when the Issue is created.
 

package/templates/.agents/rules/milestone-inference.github.zh-CN.md

@@ -1,6 +1,6 @@
 # Milestone 推断规则
 
-在 `create-issue`、`implement-task` 或 `create-pr` 处理 milestone 之前先读取本文件。
+在 `create-task` 的平台规则、`implement-task` 或 `create-pr` 处理 milestone 之前先读取本文件。
 
 ## 通用原则
 
@@ -21,7 +21,7 @@ git branch -r | grep -v 'HEAD' | grep -E 'origin/[0-9]+\.[0-9]+\.x$'
 - 有输出：多版本分支模式
 - 无输出：主干模式
 
-## 阶段 1：`create-issue`
+## 阶段 1：`create-task`（平台规则创建 Issue 时）
 
 目标：在创建 Issue 时先确定粗粒度版本线。
 

package/templates/.agents/scripts/platform-adapters/platform-sync.github.js

@@ -498,7 +498,7 @@ function checkPrCommentLastCommit(context, remoteData) {
     );
   }
 
-  const headResult = withRetry(() => gitText(["rev-parse", "HEAD"], context.taskDir));
+  const headResult = resolvePrHeadSha(context);
   if (!headResult.ok) {
     return headResult.type === "check_failed"
       ? failResult(CHECK_TYPE, headResult.message, headResult.type)
@@ -1137,6 +1137,51 @@ function gitText(args, cwd) {
   return { ok: true, value: String(result.stdout || "").trim() };
 }
 
+function resolvePrHeadSha(context) {
+  const fallback = () => withRetry(() => gitText(["rev-parse", "HEAD"], context.taskDir));
+  const branch = String(context.task?.metadata?.branch || "").trim();
+  if (!branch) {
+    return fallback();
+  }
+
+  const worktreeList = withRetry(() => gitText(["worktree", "list", "--porcelain"], context.taskDir));
+  if (!worktreeList.ok) {
+    return fallback();
+  }
+
+  const matchedWorktree = findWorktreeForBranch(worktreeList.value, branch);
+  if (!matchedWorktree) {
+    return fallback();
+  }
+
+  const headInWorktree = withRetry(() => gitText(["rev-parse", "HEAD"], matchedWorktree));
+  if (!headInWorktree.ok) {
+    return fallback();
+  }
+
+  return headInWorktree;
+}
+
+function findWorktreeForBranch(porcelainOutput, branch) {
+  let currentWorktree = "";
+  for (const rawLine of String(porcelainOutput || "").split("\n")) {
+    const line = rawLine.trimEnd();
+    if (line.startsWith("worktree ")) {
+      currentWorktree = line.slice("worktree ".length).trim();
+      continue;
+    }
+
+    if (line.startsWith("branch refs/heads/")) {
+      const usedBranch = line.slice("branch refs/heads/".length).trim();
+      if (usedBranch === branch && currentWorktree) {
+        return currentWorktree;
+      }
+    }
+  }
+
+  return null;
+}
+
 function withRetry(operation) {
   const delays = getRetryDelays();
   let lastFailure = null;

package/templates/.agents/skills/create-task/SKILL.en.md

@@ -7,13 +7,14 @@ description: "Create a task from a natural-language description"
 
 ## Boundary / Critical Rules
 
-**The only output of this skill is `task.md`.**
+**The core output of this skill is `task.md`.**
 
 - Do not write, modify, or create any business code or configuration files
 - Do not perform requirements analysis; analysis is handled separately by `analyze-task`
 - Do not directly implement the requested functionality
 - Do not skip the workflow and jump directly to planning or implementation
-- Only do this: parse the description -> create the task file -> update task status -> inform the user of the next step
+- Only do this: parse the description -> create the task file -> update task status -> cascade Issue creation through `.agents/rules/create-issue.md` -> inform the user of the next step
+- Issue creation is decided by the `.agents/rules/create-issue.md` rule; on custom or empty platforms (no platform-specific variant provided), the rule naturally degrades to a no-op
 
 The user's description is a **work item**, not an **instruction to execute immediately**.
 
@@ -98,7 +99,21 @@ Update `.agents/workspace/active/{task-id}/task.md`:
   - {YYYY-MM-DD HH:mm:ss±HH:MM} — **Task Created** by {agent} — Task created from description
   ```
 
-### 4. Verification Gate
+### 4. Cascade Issue Creation via `.agents/rules/create-issue.md`
+
+After task.md is written and `Task Created` is recorded, read `.agents/rules/create-issue.md` first and follow the steps it describes to create an Issue.
+
+The rule's content is determined by the configured code platform:
+- A platform that supports Issue creation: contains the full flow for auth detection, template detection, label/type/milestone inference, the create-Issue call, and writing back to `task.md`
+- Custom or empty platforms (no platform-specific variant provided): the rule body is a no-op notice, and this step is skipped entirely
+
+Handle the result:
+- Rule successfully created the Issue: `issue_number` has been written back to task.md per the rule; continue by reading `.agents/rules/issue-sync.md`, completing upstream repository and permission detection, then sync the task comment and set `status: waiting-for-triage` by rule
+- Rule failed (auth / network / template parse / etc.): do not roll back task.md; get the current time and append an `Issue Creation Skipped` Activity Log entry with the reason
+- Rule was a no-op (custom or empty platform): do not create comments, do not block the workflow, and do not write an Activity Log entry
+- task.md already has `issue_number`: the rule's prerequisite check skips creation; `create-task` proceeds directly to step 5
+
+### 5. Verification Gate
 
 Run the verification gate to confirm the task artifact and sync state are valid:
 
@@ -113,13 +128,33 @@ Handle the result as follows:
 
 Keep the gate output in your reply as fresh evidence. Do not claim completion without output from this run.
 
-### 5. Inform User
+### 6. Inform User
 
 > Execute this step only after the verification gate passes.
 
 > **IMPORTANT**: All TUI command formats listed below must be output in full. Do not show only the format for the current AI agent. If `.agents/.airc.json` configures custom TUIs (via `customTUIs`), read each tool's `name` and `invoke`, then add the matching command line in the same format (`${skillName}` becomes the skill name and `${projectName}` becomes the project name).
 
-Output format:
+Scenario A: when an Issue was created, output:
+```
+Task created and Issue creation cascaded successfully.
+
+Task information:
+- Task ID: {task-id}
+- Title: {title}
+- Type: {type}
+- Workflow: {workflow}
+- Issue: #{issue_number} {issue_url}
+
+Output file:
+- Task file: .agents/workspace/active/{task-id}/task.md
+
+Next step - run requirements analysis:
+  - Claude Code / OpenCode: /analyze-task {task-id}
+  - Gemini CLI: /{{project}}:analyze-task {task-id}
+  - Codex CLI: $analyze-task {task-id}
+```
+
+Scenario B: when no Issue was created, output:
 ```
 Task created.
 
@@ -132,15 +167,36 @@ Task information:
 Output file:
 - Task file: .agents/workspace/active/{task-id}/task.md
 
+Next step - run requirements analysis:
+  - Claude Code / OpenCode: /analyze-task {task-id}
+  - Gemini CLI: /{{project}}:analyze-task {task-id}
+  - Codex CLI: $analyze-task {task-id}
+```
+
+Scenario C: when Issue creation failed, output:
+```
+Task created, but cascade Issue creation failed.
+
+Task information:
+- Task ID: {task-id}
+- Title: {title}
+- Type: {type}
+- Workflow: {workflow}
+
+Issue creation failed:
+- Error code: {error_code}
+- Reason: {error_message}
+- Local task.md was kept and not rolled back
+
+Output file:
+- Task file: .agents/workspace/active/{task-id}/task.md
+
 Next step - run requirements analysis:
   - Claude Code / OpenCode: /analyze-task {task-id}
   - Gemini CLI: /{{project}}:analyze-task {task-id}
   - Codex CLI: $analyze-task {task-id}
 
-Or create an Issue first:
-  - Claude Code / OpenCode: /create-issue {task-id}
-  - Gemini CLI: /{{project}}:create-issue {task-id}
-  - Codex CLI: $create-issue {task-id}
+For later platform sync: after fixing auth / network / template issues, manually run the Issue creation flow in `.agents/rules/create-issue.md` for this task; or manually create/find an Issue and write `issue_number` into task.md so later skills can take over cascade sync.
 ```
 
 ## Completion Checklist
@@ -150,8 +206,9 @@ Or create an Issue first:
 - [ ] Updated `updated_at` to the current time in task.md
 - [ ] Updated `assigned_to` in task.md
 - [ ] Appended an Activity Log entry to task.md
+- [ ] Tried cascading Issue creation through `.agents/rules/create-issue.md`; if it failed, kept task.md and recorded the reason
 - [ ] Informed the user of the next step (must include all TUI command formats, including any custom TUIs; do not filter)
-- [ ] **Did not modify any business code or configuration files** (only task.md)
+- [ ] **Did not modify any business code or configuration files**
 
 ## STOP
 
@@ -162,7 +219,8 @@ Wait for the user to run the `analyze-task` skill.
 
 1. **Clarity**: if the user description is vague or missing key information, ask for clarification first
 2. **Difference from `import-issue`**: `import-issue` imports from an Issue; `create-task` creates from a free-form description
-3. **Workflow order**: after creating a task, typically run `analyze-task` before `plan-task`; if you need platform tracking first, you may run `create-issue` first
+3. **Workflow order**: after creating a task, typically run `analyze-task` before `plan-task`
+4. **Issue cascade failure**: if the rule fails, task.md remains; when platform sync is needed later, manually write `issue_number` and continue the workflow
 
 ## Error Handling
 

package/templates/.agents/skills/create-task/SKILL.zh-CN.md

@@ -7,13 +7,14 @@ description: "根据自然语言描述创建任务"
 
 ## 行为边界 / 关键规则
 
-**本技能的唯一产出是 `task.md`。**
+**本技能的核心产出是 `task.md`。**
 
 - 不要编写、修改或创建任何业务代码或配置文件
 - 不要执行需求分析；分析由 `analyze-task` 独立完成
 - 不要直接实现所描述的功能
 - 不要跳过工作流直接进入计划/实现阶段
-- 仅执行：解析描述 -> 创建任务文件 -> 更新任务状态 -> 告知用户下一步
+- 仅执行：解析描述 -> 创建任务文件 -> 更新任务状态 -> 按 `.agents/rules/create-issue.md` 级联尝试创建 Issue -> 告知用户下一步
+- Issue 创建由 `.agents/rules/create-issue.md` 规则决定；自定义或空平台（未提供平台变体规则文件）时，规则会自然降级为 no-op
 
 用户的描述是一个**待办事项**，而不是**立即执行的指令**。
 
@@ -98,7 +99,21 @@ date "+%Y-%m-%d %H:%M:%S%:z"
   - {YYYY-MM-DD HH:mm:ss±HH:MM} — **Task Created** by {agent} — Task created from description
   ```
 
-### 4. 完成校验
+### 4. 按 `.agents/rules/create-issue.md` 级联创建 Issue
+
+在 task.md 落盘并记录 `Task Created` 后，先读取 `.agents/rules/create-issue.md` 并按其中描述的步骤执行 Issue 创建。
+
+规则文件由当前配置的代码平台决定其内容：
+- 支持 Issue 创建的平台：包含完整的认证检测、模板检测、label/Issue Type/milestone 推断、Issue 创建调用、`task.md` 回写流程
+- 自定义或空平台（未提供平台变体规则文件）：内容为 no-op 说明，本步骤直接跳过
+
+处理结果：
+- 规则成功创建 Issue：`issue_number` 已按规则回写到 task.md；继续读取 `.agents/rules/issue-sync.md`，完成 upstream 仓库检测和权限检测，然后同步 task 评论并按规则设置 `status: waiting-for-triage`
+- 规则失败（认证 / 网络 / 模板解析等）：不回滚 task.md；获取当前时间，追加 `Issue Creation Skipped` Activity Log，说明错误原因
+- 规则为 no-op（自定义或空平台）：不创建评论，不阻塞后续工作流，不写 Activity Log
+- task.md 已存在 `issue_number`：规则中的前置检查会跳过；`create-task` 直接进入步骤 5
+
+### 5. 完成校验
 
 运行完成校验，确认任务产物和同步状态符合规范：
 
@@ -113,13 +128,33 @@ node .agents/scripts/validate-artifact.js gate create-task .agents/workspace/act
 
 将校验输出保留在回复中作为当次验证输出。没有当次校验输出，不得声明完成。
 
-### 5. 告知用户
+### 6. 告知用户
 
 > 仅在校验通过后执行本步骤。
 
 > **重要**：以下「下一步」中列出的所有 TUI 命令格式必须完整输出，不要只展示当前 AI 代理对应的格式。如果 `.agents/.airc.json` 中配置了自定义 TUI（`customTUIs`），读取每个工具的 `name` 和 `invoke`，按同样格式补充对应命令行（`${skillName}` 替换为技能名，`${projectName}` 替换为项目名）。
 
-输出格式：
+场景 A：已创建 Issue 时输出：
+```
+任务已创建，并已级联创建 Issue。
+
+任务信息：
+- 任务 ID：{task-id}
+- 标题：{title}
+- 类型：{type}
+- 工作流：{workflow}
+- Issue：#{issue_number} {issue_url}
+
+产出文件：
+- 任务文件：.agents/workspace/active/{task-id}/task.md
+
+下一步 - 执行需求分析：
+  - Claude Code / OpenCode：/analyze-task {task-id}
+  - Gemini CLI：/{{project}}:analyze-task {task-id}
+  - Codex CLI：$analyze-task {task-id}
+```
+
+场景 B：未创建 Issue 时输出：
 ```
 任务已创建。
 
@@ -134,13 +169,34 @@ node .agents/scripts/validate-artifact.js gate create-task .agents/workspace/act
 
 下一步 - 执行需求分析：
   - Claude Code / OpenCode：/analyze-task {task-id}
-  - Gemini CLI：/agent-infra:analyze-task {task-id}
+  - Gemini CLI：/{{project}}:analyze-task {task-id}
+  - Codex CLI：$analyze-task {task-id}
+```
+
+场景 C：Issue 创建失败时输出：
+```
+任务已创建，但 Issue 级联创建失败。
+
+任务信息：
+- 任务 ID：{task-id}
+- 标题：{title}
+- 类型：{type}
+- 工作流：{workflow}
+
+Issue 创建失败：
+- 错误码：{error_code}
+- 原因：{error_message}
+- 本地 task.md 已保留，未回滚
+
+产出文件：
+- 任务文件：.agents/workspace/active/{task-id}/task.md
+
+下一步 - 执行需求分析：
+  - Claude Code / OpenCode：/analyze-task {task-id}
+  - Gemini CLI：/{{project}}:analyze-task {task-id}
   - Codex CLI：$analyze-task {task-id}
 
-或先创建 Issue：
-  - Claude Code / OpenCode：/create-issue {task-id}
-  - Gemini CLI：/agent-infra:create-issue {task-id}
-  - Codex CLI：$create-issue {task-id}
+后续如需平台同步：修复认证/网络/模板问题后，可按 `.agents/rules/create-issue.md` 对当前任务手动执行一次 Issue 创建；或手动创建/查找 Issue，并把 `issue_number` 写入 task.md，后续技能会接管级联同步。
 ```
 
 ## 完成检查清单
@@ -150,8 +206,9 @@ node .agents/scripts/validate-artifact.js gate create-task .agents/workspace/act
 - [ ] 更新了 task.md 中的 `updated_at` 为当前时间
 - [ ] 更新了 task.md 中的 `assigned_to`
 - [ ] 追加了 Activity Log 条目到 task.md
+- [ ] 已按 `.agents/rules/create-issue.md` 尝试级联创建 Issue；失败时保留 task.md 并记录原因
 - [ ] 告知了用户下一步（必须展示所有 TUI 的命令格式，含自定义 TUI，不要筛选）
-- [ ] **没有修改任何业务代码或配置文件**（仅 task.md）
+- [ ] **没有修改任何业务代码或配置文件**
 
 ## 停止
 
@@ -162,7 +219,8 @@ node .agents/scripts/validate-artifact.js gate create-task .agents/workspace/act
 
 1. **清晰度**：如果用户描述模糊或缺少关键信息，先要求澄清
 2. **与 import-issue 的区别**：`import-issue` 从 Issue 导入任务；`create-task` 从自由描述创建
-3. **工作流顺序**：创建任务后，通常先执行 `analyze-task` 再进入 `plan-task`；如需先建立 平台跟踪，也可先执行 `create-issue`
+3. **工作流顺序**：创建任务后，通常先执行 `analyze-task` 再进入 `plan-task`
+4. **Issue 级联失败**：如果规则执行失败，task.md 仍保留；需要后续平台同步时，可手动写入 `issue_number` 后继续执行工作流
 
 ## 错误处理
 

package/templates/.agents/skills/create-task/config/verify.json

@@ -20,6 +20,11 @@
       "expected_action_pattern": "Task Created",
       "freshness_minutes": 30
     },
-    "platform-sync": null
+    "platform-sync": {
+      "when": "issue_number_exists",
+      "issue_must_exist": true,
+      "verify_task_comment_content": true,
+      "expected_status_label_key": "waitingForTriage"
+    }
   }
 }

package/templates/.agents/skills/implement-task/reference/implementation-rules.en.md

@@ -16,31 +16,26 @@ Follow the `implementation` step in `.agents/workflows/feature-development.yaml`
 **Implementation principles**:
 1. **Follow the plan strictly**: do not deviate from the technical plan
 2. **Work step by step**: execute the planned sequence
-3. **Keep testing continuously**: re-run tests as work progresses
+3. **Keep testing continuously**: run the **smoke subset** continuously as work progresses (see the `test` skill)
 4. **Keep it simple**: do not add unplanned features
 
 ## Run Test Verification
 
-Run the project test command. Refer to the `test` skill for project-specific commands.
+During implementation:
+- **Inner loop**: after each change, run the project's **smoke subset** (see the `test` skill) for fast feedback
+- **Before writing the implementation report**: run the **core subset** as final verification so code entering review has passed the complete core checks
 
-```bash
-# See .agents/skills/test/SKILL.md for the project's test command
-# Common patterns:
-# npm test          (Node.js)
-# mvn test          (Maven)
-# pytest            (Python)
-# go test ./...     (Go)
-```
+> Refer to the `test` skill for project-specific commands; downstream projects without layered scripts should fall back to the full project test command.
 
 If tests fail:
 - analyze the failure first and prioritize fixing issues introduced by this implementation
-- re-run tests after each fix until they pass, or confirm that the problem is an external blocker
+- after each fix, re-run at least the smoke subset, then upgrade to core for the next full-pass verification
 - only stop without producing the implementation artifact when the failure is caused by an external blocker, missing environment, or unclear requirement that cannot be resolved inside the task
 
 Two-way failure handling:
 1. implementation-caused failures:
    - fix the code, tests, docs, or fixtures introduced by this implementation
-   - re-run the test command after each fix
+   - re-run tests after each fix (smoke for the immediate fix verification, core for the round-level verification)
    - continue until all required tests pass
 2. external blockers:
    - confirm the failure comes from missing environment, unrelated upstream breakage, or unclear requirements outside this task

package/templates/.agents/skills/implement-task/reference/implementation-rules.zh-CN.md

@@ -16,31 +16,26 @@
 **实现原则**：
 1. **严格遵循方案**：不要偏离技术方案
 2. **按步骤推进**：按计划顺序执行
-3. **持续测试**：实现过程中持续重新运行测试
+3. **持续测试**：实现过程中持续运行 **smoke 子集**（参见 `test` skill）
 4. **保持简单**：不要加入未计划的功能
 
 ## 运行测试验证
 
-运行项目测试命令。项目特定命令请参考 `test` skill。
+实现过程中：
+- **内循环**：每次修改后运行项目测试的 **smoke 子集**（参见 `test` skill），快速反馈
+- **写实现报告前**：运行 **core 子集**做最终验证，确保进入审查的代码已通过完整的核心测试
 
-```bash
-# 参考 .agents/skills/test/SKILL.md 中的项目测试命令
-# 常见模式：
-# npm test          (Node.js)
-# mvn test          (Maven)
-# pytest            (Python)
-# go test ./...     (Go)
-```
+> 项目特定命令请参考 `test` skill；不存在分层 script 的下游项目回退到完整项目测试命令。
 
 如果测试失败：
 - 先分析失败原因，并优先修复由本次实现引入的问题
-- 每次修复后重新运行测试，直到通过，或者确认问题属于外部阻塞
+- 每次修复后重新运行至少 smoke 子集，进入下一轮再升级到 core
 - 只有当失败由外部阻塞、环境缺失或当前任务内无法澄清的需求问题导致时，才允许在不生成实现产物的情况下停止
 
 双分支失败处理：
 1. 本次实现引入的失败：
    - 修复由本次实现引入的代码、测试、文档或 fixture 问题
-   - 每次修复后重新运行测试命令
+   - 每次修复后重新运行测试（smoke 用于本次修复验证，core 用于本轮整体验证）
    - 持续处理直到所有必需测试通过
 2. 外部阻塞：
    - 确认失败来自缺失环境、无关的上游损坏，或任务范围外的需求不明确

package/templates/.agents/skills/import-issue/SKILL.en.md

@@ -108,7 +108,7 @@ If task.md contains a valid `issue_number`, use the Issue update command from `.
 
 If task.md contains a valid `issue_number`, perform these sync actions (skip and continue on any failure):
 - Read `.agents/rules/issue-sync.md` before syncing, and complete upstream repository detection plus permission detection
-- Check the Issue's current milestone; if it is unset, read `.agents/rules/milestone-inference.md` and infer plus set the milestone using "Stage 1: `create-issue`". If `has_triage=false` or the inference is uncertain, skip and continue
+- Check the Issue's current milestone; if it is unset, read `.agents/rules/milestone-inference.md` and infer plus set the milestone using "Phase 1: `create-task` (when the platform rule creates an Issue)". If `has_triage=false` or the inference is uncertain, skip and continue
 - After every scenario, task comment sync is mandatory: create or update the task comment marker defined in `.agents/rules/issue-sync.md` so the remote `:task` comment exists and matches the local `task.md` content (follow the task.md comment sync rule in issue-sync.md)
 
 ### 7. Verification Gate

package/templates/.agents/skills/import-issue/SKILL.zh-CN.md

@@ -108,7 +108,7 @@ date "+%Y-%m-%d %H:%M:%S%:z"
 
 如果 task.md 中存在有效的 `issue_number`，执行以下同步操作（任一失败则跳过并继续）：
 - 执行前先读取 `.agents/rules/issue-sync.md`，完成 upstream 仓库检测和权限检测
-- 检查 Issue 当前 milestone；如果未设置，先读取 `.agents/rules/milestone-inference.md`，按其中的「阶段 1：`create-issue`」规则推断并设置 milestone；如果 `has_triage=false` 或推断不确定，跳过并继续
+- 检查 Issue 当前 milestone；如果未设置，先读取 `.agents/rules/milestone-inference.md`，按其中的「阶段 1：`create-task`（平台规则创建 Issue 时）」规则推断并设置 milestone；如果 `has_triage=false` 或推断不确定，跳过并继续
 - 所有场景结束后，必须执行一次 task 留言同步，创建或更新 `.agents/rules/issue-sync.md` 中定义的 task 评论标记，确保远端 `:task` 评论存在且内容与本地 `task.md` 一致（按 issue-sync.md 的 task.md 评论同步规则）
 
 ### 7. 完成校验

package/templates/.agents/skills/refine-task/reference/fix-workflow.en.md

@@ -26,11 +26,11 @@ For each fix:
 1. read the affected files
 2. apply the smallest necessary change
 3. verify the change addresses the review feedback
-4. run the relevant tests
+4. run the project's **smoke subset** for immediate feedback (see the `test` skill)
 
 ## Run Test Verification
 
-Run the project test command from the `test` skill and confirm that all required tests still pass.
+Before writing the refinement report, run the project's **core subset** as final verification and confirm that all required tests still pass. If the project does not have layered scripts, fall back to the full project test command.
 
 ## Choose the Next-Step Branch
 

package/templates/.agents/skills/refine-task/reference/fix-workflow.zh-CN.md

@@ -26,11 +26,11 @@
 1. 读取受影响文件
 2. 施加最小必要改动
 3. 验证改动确实解决了审查反馈
-4. 运行相关测试
+4. 运行项目测试的 **smoke 子集**做即时反馈（参见 `test` skill）
 
 ## 运行测试验证
 
-运行项目测试命令。项目测试命令以 `test` skill 为准，并确认所有必需测试仍然通过。
+写 refinement 报告前，运行项目测试的 **core 子集**做最终验证，确保所有必需测试仍然通过。如果项目没有分层 script，回退到完整项目测试命令。
 
 ## 选择下一步分支
 

package/templates/.agents/skills/test/SKILL.en.md

@@ -21,16 +21,55 @@ Execute the project's full test workflow including compilation checks and unit t
 
 Confirm no compilation errors.
 
-## 2. Run All Unit Tests
+## 2. Run Unit Tests by Layer
+
+This project uses three test layers as an optional optimization; if the test suite is small, all layers may map to the same full test command.
+
+### smoke (target <5s)
+
+```bash
+# TODO: Replace with this project's smoke subset command
+# npm run test:smoke       (Node.js)
+# pytest -m "not slow"     (Python)
+# go test -short ./...     (Go)
+```
+
+Use for:
+- implement-task / refine-task inner loops
+- save-and-run / frequent feedback
+- project structure, configuration, and template contract checks
+
+### core (target <15s)
 
 ```bash
-# TODO: Replace with your project's test command
-# npm test                (Node.js)
-# mvn test                (Maven)
-# pytest                  (Python)
-# go test ./...           (Go)
+# TODO: Replace with this project's core subset command
+# npm run test:core        (Node.js)
+# pytest -m "not contract" (Python)
+# go test ./...            (Go)
 ```
 
+Use for:
+- pre-commit hook (automatic)
+- final verification before writing implementation.md / refinement.md
+- local gate before pushing a PR
+
+### full (complete test suite)
+
+```bash
+# TODO: Replace with this project's full test command
+# npm test                 (Node.js)
+# mvn test                 (Maven)
+# pytest                   (Python)
+# go test ./...            (Go)
+```
+
+Use for:
+- before release / tag
+- CI
+- final gate before merging to main
+
+If the project is not layered yet, smoke / core / full may all map to the same full test command; layering is a feedback-speed optimization, not a prerequisite for using the collaboration workflow.
+
 ## 3. Output Results
 
 Report test result summary:

package/templates/.agents/skills/test/SKILL.zh-CN.md

@@ -21,16 +21,55 @@ description: "执行项目完整测试流程"
 
 确认无编译错误。
 
-## 2. 运行所有单元测试
+## 2. 运行单元测试（按层级选择）
+
+本项目把测试分为三层（可选优化）；如果项目测试规模较小，可以全部映射到同一个完整测试命令。
+
+### smoke（目标 <5s）
+
+```bash
+# TODO: 替换为本项目的 smoke 子集命令
+# npm run test:smoke       (Node.js)
+# pytest -m "not slow"     (Python)
+# go test -short ./...     (Go)
+```
+
+适用场景：
+- implement-task / refine-task 内循环
+- 保存即跑 / 频繁反馈
+- 仅断言项目结构、配置、模板契约
+
+### core（目标 <15s）
 
 ```bash
-# TODO: 替换为你的项目测试命令
-# npm test                (Node.js)
-# mvn test                (Maven)
-# pytest                  (Python)
-# go test ./...           (Go)
+# TODO: 替换为本项目的 core 子集命令
+# npm run test:core        (Node.js)
+# pytest -m "not contract" (Python)
+# go test ./...            (Go)
 ```
 
+适用场景：
+- pre-commit hook（自动调用）
+- 写 implementation.md / refinement.md 报告前的最终验证
+- 推送 PR 前的本地把关
+
+### full（完整测试）
+
+```bash
+# TODO: 替换为本项目的完整测试命令
+# npm test                 (Node.js)
+# mvn test                 (Maven)
+# pytest                   (Python)
+# go test ./...            (Go)
+```
+
+适用场景：
+- release / tag 前
+- CI
+- main 合并前的最终把关
+
+如果项目暂不分层，smoke / core / full 可以全部映射到同一个完整测试命令；分层是反馈速度优化项，不是使用协作工作流的前置条件。
+
 ## 3. 输出结果
 
 报告测试结果摘要：