@boyingliu01/xp-gate 0.7.0 → 0.8.0

package/skills/test-specification-alignment/SKILL.md

@@ -1,10 +1,20 @@
 ---
 name: test-specification-alignment
-description: "测试与 Specification 对齐验证引擎。确保测试准确反映需求和设计。两阶段执行：Phase 1 验证对齐（可修改测试），Phase 2 执行测试（禁止修改测试）。MANDATORY before any release. TRIGGER: 'run tests', 'verify tests', before BUILD (TDD + review) Arbiter, before gstack-ship."
+description: "Use when asked to run tests, verify tests, align tests with specification.yaml, before BUILD verification, or before release/ship."
 ---
 
 # Test-Specification Alignment Engine
 
+## Workflow Steps
+
+1. **Load Phase 0** — Verify `specification.yaml` and `tests/` directory exist; BLOCK if missing with clear guidance to complete brainstorming → delphi-review → specification-generator flow first
+2. **Phase 1 Alignment (Modify Allowed)** — Parse specification.yaml (YAML) + test files (AST); validate alignment rules; generate alignment report; optionally fix tests to align with requirements; require score ≥80% to proceed
+3. **Coverage Mapping** — Map each REQ-* to at least one test case with `@test` annotation; map each AC-* to at least one assertion; verify test intent declarations (`@test`, `@intent`, `@covers` tags)
+4. **Pre-Phase 2 Freeze** — Invoke `/freeze` skill to lock test directories (`tests/`, `test/`, `__tests__/`, `*.test.ts`, `*.spec.ts`, `*_test.py`, etc.); return confirmation before proceeding
+5. **Phase 2 Execution (Frozen)** — Run all tests; Agent is BLOCKED from modifying/deleting/skipping tests; analyze failures into 4 categories: BUSINESS_CODE_ERROR, TEST_DATA_ERROR, SPECIFICATION_ERROR, ENVIRONMENT_ERROR; SPECIFICATION_ERROR requires ESCALATE_TO_HUMAN
+6. **Failure Classification** — For each failing test, output classification in mandatory format with Type, Test, Root Cause, Action fields; if SPECIFICATION_ERROR detected, offer user options (A: fix spec → re-Phase 1, B: confirm spec → modify code, C: clarify ambiguity)
+7. **JSON Report Output** — Generate alignment report as valid JSON with fields: `alignment_status` (PASS|FAIL|BLOCKED), `phase` (1|2), `score`, `misaligned_tests[]`, `anti_pattern_detected`, `errors[]`; Terminal State: ✅ ALL_TESTS_PASS
+
 ## 核心原则
 
 **测试是系统的防护网，也是系统的使用手册。测试必须准确反映原始需求和设计方案。**
@@ -37,6 +47,89 @@ Alignment report MUST be output as valid JSON:
 
 ---
 
+## Triggers
+
+### Automatic Triggers
+- BUILD (TDD + review) Round 1 after Driver outputs tests
+- Gate 1 verification before proceeding
+- gstack-ship release before deployment
+
+### Manual Triggers
+- `/test-specification-alignment` command
+- `/verify-tests` command
+- "run tests" phrase
+- "verify tests" phrase
+- before gstack-ship
+
+---
+
+## Scope
+
+### IN Scope
+- Validation of test alignment against `specification.yaml` (Requirements, User Stories, Acceptance Criteria)
+- Phase 1: Test modification to improve alignment (add missing tests, fix annotations, correct intent)
+- Phase 2: Test execution with frozen test files (no modifications allowed)
+- Failure classification into BUSINESS_CODE_ERROR, TEST_DATA_ERROR, SPECIFICATION_ERROR, ENVIRONMENT_ERROR
+- JSON report generation with alignment status, score, misaligned tests, and errors
+
+### OUT Scope
+- Modifying `specification.yaml` during Phase 2 (requires ESCALATE_TO_HUMAN)
+- Skipping or deleting tests to make tests pass
+- Modifying test assertions to force tests to pass
+- Business logic implementation (handled by BUILD phase)
+- Environment configuration issues (handled separately)
+
+### Boundaries
+- **Start**: After BUILD (TDD + review) Round 1 outputs tests
+- **End**: Terminal State ✅ ALL_TESTS_PASS or ESCALATE_TO_HUMAN for specification issues
+- **Inputs**: `specification.yaml`, `tests/` directory
+- **Outputs**: Alignment report (JSON), test execution report, failure classification
+
+---
+
+## Examples
+
+### Example 1: Normal Trigger (Should Execute)
+**User Input**: "run tests for login feature" or `/test-specification-alignment`
+
+**Expected Flow**:
+1. Load Phase 0: Check `specification.yaml` exists → ✅ Found
+2. Check `tests/` directory exists → ✅ Found
+3. Phase 1: Parse spec and tests → Alignment score 85% → ✅ Pass threshold
+4. Pre-Phase 2: Call `/freeze tests/` → ✅ Locked
+5. Phase 2: Run tests → 12 passed, 0 failed → ✅ ALL_TESTS_PASS
+6. Output JSON report with `alignment_status: "PASS"`, `phase: "2"`, `score: 85`
+7. Call `/unfreeze tests/` → ✅ Unlocked
+8. Terminal State: ✅ Test-specification-alignment complete
+
+---
+
+### Example 2: Should NOT Trigger (Missing Specification)
+**User Input**: "run tests" in a project without `specification.yaml`
+
+**Expected Flow**:
+1. Load Phase 0: Check `specification.yaml` exists → ❌ Not Found
+2. BLOCK with message:
+   ```
+   ❌ BLOCKED: specification.yaml 不存在
+
+   要生成 specification.yaml，请先完成需求流程：
+
+   流程步骤:
+   1. brainstorming      → 需求探索，生成设计文档
+   2. delphi-review      → 需求评审 (多轮直到 APPROVED)
+   3. spec 自动生成 (从 APPROVED 设计文档提取 specification.yaml)
+
+   为什么必须这样？
+   - 在 delphi-review APPROVED 后生成，避免设计文档修改时 spec 也需重新生成
+   - 遵循"问题发现越早修复成本越低"原则
+
+   请先完成上述流程，然后再运行 test-specification-alignment。
+   ```
+3. Terminal State: ❌ BLOCKED_MISSING_SPECIFICATION
+
+---
+
 ## 触发条件
 
 ### 自动触发
@@ -569,29 +662,6 @@ legacy_mode:
 
 ---
 
-## Qoder 平台适配
-
-### freeze 机制替代（替代 gstack/freeze skill）
-
-在 Qoder 环境中，freeze/unfreeze 通过 sprint-flow 的 **Pre-Edit Gate** 替代：
-- Phase 2 执行期间，orchestrator 禁止修改测试目录下的文件
-- 此约束通过 SKILL.md 指令强制执行（非物理阻断）
-- orchestrator 在每次文件编辑前检查：如果目标文件位于测试目录且当前处于 Phase 2 freeze 状态，则 **BLOCK**
-
-### Agent 配置适配
-
-| 原配置 | Qoder 替代 | 说明 |
-|---------|------------|------|
-| Phase 1 Agent (Qwen3.5-Plus) | **orchestrator 直接执行** | YAML 解析 + AST 解析由 orchestrator 内联完成 |
-| Phase 2 Agent (GLM-5) | **orchestrator 直接执行** | 测试执行由 orchestrator 通过 Bash 工具完成 |
-
-### Qoder 集成点
-
-- Phase 3 REVIEW 完成后，使用 genui `show_widget` 展示对齐报告摘要
-- 对齐分数记录通过 `UpdateMemory` 持久化（development_test_specification 类型）
-
----
-
 ## Anti-Patterns
 
 | 错误 | 正确 |
@@ -611,6 +681,10 @@ legacy_mode:
 |------|------|------|
 | V1.0 | 2026-04-06 | 初始设计 |
 | V2.0 | 2026-04-06 | Delphi Review 共识版本 |
+| V2.1 | 2026-06-02 | Added parser-friendly sections: Workflow Steps (7 steps), Triggers (5+ phrases), Scope (IN/OUT/Boundaries), Examples (2 scenarios) |
+
+---
+
 ## Output Format (MANDATORY)
 Alignment report MUST be output as valid JSON:
 ```json

package/skills/to-issues/SKILL.md

@@ -0,0 +1,277 @@
+---
+name: to-issues
+description: Break a plan, spec, or PRD into independently-grabbable issues using tracer-bullet vertical slices. Use when user wants to convert a plan into issues, create implementation tickets, or break down work into vertical slices. Integrates with XP-Gate sprint-flow Phase 1.
+---
+
+# To Issues — Vertical Slice Breakdown
+
+将 PRD / specification.yaml / 设计文档拆解为 **独立的垂直切片 Issue**，每个切片是贯穿所有层的 **tracer bullet**，而非水平分层切片。
+
+核心理念：每个 Issue 交付一条**狭窄但完整**的端到端路径（schema → API → UI → tests），完成后即可独立演示并通过质量门禁。
+
+---
+
+## 术语（强制一致使用）
+
+| 术语 | 定义 |
+|------|------|
+| **Vertical Slice** | 贯穿所有层（数据、API、UI、测试）的薄路径，交付端到端行为 |
+| **Tracer Bullet** | 第一条完整的垂直切片，优先拆解以快速暴露 unknown unknowns |
+| **HITL** | Human-In-The-Loop — 需要人工决策（架构选择、设计评审）才能继续 |
+| **AFK** | Away-From-Keyboard — 可完全由 agent 自主实现并合并 |
+| **Blocking** | 依赖关系：此切片必须在哪些切片完成后才能开始 |
+| **Deep Module** | 封装大量功能于简单接口背后的模块（XP-Gate Principles 概念） |
+
+---
+
+## 输入
+
+- PRD 文件路径，或
+- `specification.yaml`，或
+- 最近对话中的设计文档（项目配置的 spec 路径，默认 `docs/superpowers/specs/*.md`）
+
+## 输出（Machine-Readable）
+
+### 1. 文本输出
+
+- 一组垂直切片 Issue（GitHub Issue / 本地 `.issues/*.md` / 用户指定的 issue tracker 格式）
+- 依赖关系图（拓扑排序后的执行顺序）
+
+### 2. JSON Output Contract（供 ralph-loop / sprint-flow 下游消费）
+
+写入 `.sprint-state/slices-manifest.json`：
+
+```json
+{
+  "version": 1,
+  "source": "specification.yaml 或 spec 文件路径",
+  "slices": [
+    {
+      "id": "slice-N",
+      "title": "一句话描述端到端行为",
+      "type": "HITL | AFK",
+      "blocked_by": ["slice-M"] | [],
+      "parent_story": "User Story 编号",
+      "what_it_delivers": "端到端行为描述",
+      "acceptance_criteria": ["标准1", "标准2", "标准3"],
+      "effort": "S | M | L",
+      "loc_budget": 200,
+      "file_budget": 20,
+      "status": "pending"
+    }
+  ],
+  "execution_order": ["slice-1", "slice-2", "slice-3"],
+  "dependency_graph": {
+    "slice-1": [],
+    "slice-2": ["slice-1"],
+    "slice-3": ["slice-1"]
+  }
+}
+```
+
+---
+
+## 流程
+
+### 1. 收集上下文
+
+从当前对话或输入的 PRD/spec 中提取：
+- 需求列表 / User Stories
+- 系统模块 / 架构设计
+- 技术约束（语言、框架、数据库）
+
+如果输入中包含 CONTEXT.md 或 ADRs，读取并尊重其中的领域术语和架构决策。
+
+**PRD 完整性检查（降级规则）**：
+```
+IF PRD 缺少关键内容：
+  - 缺少数据模型 / schema 定义 → 标注所有数据相关切片为 HITL
+  - 缺少 API 契约 → 标注所有 API 相关切片为 HITL
+  - 缺少 UI 设计 / 交互说明 → 标注所有 UI 相关切片为 HITL
+  - 缺少验收标准 → 标注整体验证切片为 HITL
+```
+
+### 2. 探索代码库
+
+**Brownfield 项目（已存在代码）— 必须执行**：
+- 读取 AGENTS.md CODE MAP + 目录结构
+- 检查已有模块边界、路由定义、DB 模式
+- 确认测试覆盖现状
+
+**Greenfield 项目（新项目）— 可选**：
+- 若为全新项目，跳过此步骤
+- 建议采用 schema-first 切片顺序（先 DB schema → 再 API → 再 UI）
+
+### 3. 起草垂直切片
+
+将需求拆解为 **tracer bullet** 切片。规则：
+
+- 每个切片交付**一条狭窄但完整**的端到端路径
+- **优先 unknown unknowns**：优先拆解能快速暴露不确定因素的切片
+- **切片必须可独立演示并完整通过 6 道质量门禁**（Code Quality + Dup Code + Complexity + Principles + Tests + Architecture）
+- **标注依赖**：明确哪些切片阻塞哪些切片
+- **识别 Deep Module 机会**：拆解时主动提出可封装为 deep module 的模块边界
+
+**范围控制（硬上限）**：
+| 约束 | 阈值 | 超限处理 |
+|------|------|---------|
+| 切片总数 | ≤15 / 功能模块 | 超过 → 按 epic 分组，组内拓扑排序 |
+| 单切片变更量 | ≤200 LOC 或 ≤20 文件 | 超过 → 拆分为子切片 |
+| 单切片工作量 | ≤4h 开发时间 | 超过 → 拆分为子切片 |
+
+### 4. HITL / AFK 分类
+
+**判定规则**（满足任一条件即为对应类型）：
+
+```
+HITL 条件：
+  - 涉及架构决策（数据库选型、API 风格、auth 方案）
+  - 涉及 UI/UX 设计选择（布局、交互、品牌色）
+  - 涉及第三方服务集成选型（支付、短信、OAuth provider）
+  - PRD 降级规则触发（见步骤 1）
+
+AFK 条件（必须同时满足全部）：
+  - 纯实现、遵循项目中已有模式（如已有 adapter 的 CRUD）
+  - 接口契约已在 spec/PRD 中明确定义
+  - 无新增外部依赖（第三方 API、新库引入）
+  - 测试策略可复用已有测试适配器
+
+否则 → HITL（保守默认）
+```
+
+**HITL → AFK 状态机**：
+```
+HITL (创建) → 人工确认 (ADRs / 设计决策产出) → 标记为 AFK-ready → AFK (agent 执行)
+                                                                   ↓
+                                         AFK 执行遇到未预见依赖 → 回退 HITL → BLOCKED
+```
+
+### 5. 依赖验证
+
+在发布前执行以下检查：
+
+```
+1. 构建依赖图 (DAG)
+2. 检测循环依赖 (A→B→C→A) → 报错并要求重新拆分
+3. 拓扑排序 → 生成 execution_order
+4. 检查每个切片的 blocked_by 引用是否存在
+5. 验证第一个切片（tracer bullet）无阻塞
+```
+
+**跨切面关注点处理**：
+Auth、logging、error handling、i18n 等贯穿所有切层的关注点：
+- **首选**：在第一条 tracer bullet（通常是 slice-1）中内置这些横切关注点的基础设施
+- **次选**：作为独立的前置 AFK 切片（如 "slice-0: 基础设施骨架"）
+- 不要将这些关注点分散到每个切片中 — 会导致重复实现和不一致
+
+### 6. 向用户确认
+
+以编号列表呈现拆解结果。每个切片包含：
+
+```
+### Slice N: [简短标题]
+- **Type**: HITL / AFK
+- **Effort**: S(~1h) / M(~2h) / L(~4h)
+- **Blocked by**: Slice M / None
+- **Parent Story**: User Story 编号
+- **LOC Budget**: ~N 行
+- **What it delivers**: 一句话描述端到端行为
+- **Acceptance Criteria**: 1-3 条可验证行为
+```
+
+询问用户：
+1. 粒度是否合适？（太粗 / 太细 / 正好）
+2. 依赖关系是否正确？
+3. 是否有切片应该合并或拆分？
+4. HITL/AFK 标注是否准确？
+
+**最多 2 轮迭代**。超过 2 轮未决则强制收敛：
+- 第一轮：完整列表展示
+- 后续轮次：仅展示变更部分（Diff 视图：新增 / 合并 / 拆分项对比）
+
+### 7. 发布切片
+
+用户批准后，生成切片 Issue 并写入 `slices-manifest.json`。
+
+**GitHub Issue 格式**:
+```markdown
+## Parent
+[父 issue 编号或 PRD 链接]
+
+## Slice ID
+slice-N
+
+## What to build
+[端到端行为描述，非逐层实现描述]
+
+## Acceptance criteria
+- [ ] 标准 1（端到端行为可验证）
+- [ ] 标准 2
+- [ ] 标准 3
+
+## Technical Notes
+[可选：接口设计、schema 变更、架构决策]
+
+## Blocked by
+[#N - 阻塞 issue 编号 / None — 可以立即开始]
+
+## Type / Effort
+[HITL | AFK] / [S | M | L]
+
+## Gate-Compliant DoD ⚠️
+此切片完成后必须通过全部 6 道质量门禁：
+- [ ] Gate 1: Code Quality (lint 无错误)
+- [ ] Gate 2: Dup Code (≤5% 相似度)
+- [ ] Gate 3: Complexity (≤5 警告)
+- [ ] Gate 4: Principles (Clean Code + SOLID 零错误)
+- [ ] Gate 5: Tests (全部通过 + ≥80% 覆盖率)
+- [ ] Gate 6: Architecture (无新增警告 + 童子军规则)
+```
+
+**按拓扑排序顺序发布**：先发布无阻塞的切片，再发布被阻塞的切片。
+
+---
+
+## 与 XP-Gate Sprint Flow 集成
+
+完整 7 阶段流程中 to-issues 的位置及切片在各阶段的生命周期：
+
+```
+Phase 0: THINK    → brainstorming → 设计文档 → APPROVED
+Phase 1: PLAN     → autoplan → delphi-review → specification.yaml
+                  → **/to-issues** → slices-manifest.json + Issue 列表
+Phase 2: BUILD    → 按 execution_order 逐个切片执行
+                  → 每个切片 = 一个 REQ（ralph-loop 模式）
+                  → 或通过 dispatching-parallel-agents 并行（无依赖的 AFK 切片）
+Phase 3: REVIEW   → **整体 review**（全部切片完成后统一 code-walkthrough）
+                  → browse / QA 验证端到端行为
+Phase 4: USER ACC → 人工验收（全部切片集成后的完整功能）
+Phase 5: FEEDBACK → learn + retro（按切片统计通过率/返工率）
+Phase 6: SHIP     → 全部切片合并后一次性 merge（或按 epic 分批 merge）
+```
+
+**切片与 REQ 的映射**（ralph-loop 模式）：
+```
+每个切片 → 生成一个 REQ 实例：
+  REQ-N:
+    - source_slice: slice-N
+    - input: slices-manifest.json 中对应切片的 acceptance_criteria
+    - constraints: LOC_budget, file_budget, type (HITL/AFK)
+    - output: 通过 6 gates 的增量 diff
+    - verification: test-specification-alignment + gate verification
+```
+
+---
+
+## Anti-Patterns
+
+| ❌ 错误 | ✅ 正确 |
+|---------|---------|
+| 按技术层拆分（"先做 DB"、"再做 API"、"最后做 UI"） | 按端到端行为拆分（"用户可以通过邮箱注册"） |
+| 伪装成垂直切片的水平拆分（标题是用户故事，执行时仍分层） | 每个切片的 acceptance criteria 必须包含全层验证 |
+| 切片完成后无法独立演示 | 每个切片必须可验证 + 通过 6 道质量门禁 |
+| 一个切片包含过多用户故事（>2 个） | 一个切片聚焦 1-2 个用户故事 |
+| 忽略或随意标注依赖关系 | 拓扑排序验证 + 循环检测 |
+| HITL 切片被当作 AFK 执行 | 人工输出 ADR/决策后 → 标记 AFK-ready → 执行 |
+| 切片标题模糊（"实现功能X"） | 切片标题描述行为（"用户可通过邮箱注册账号"） |
+| 切片超出 200 LOC / 20 文件 / 4h 预算 | 拆分为子切片 |

package/plugins/qoder/AGENTS.md

@@ -1,93 +0,0 @@
-# QODER PLUGIN KNOWLEDGE BASE
-
-**Generated:** 2026-06-03
-**Version:** v0.6.0
-
-## OVERVIEW
-Cross-platform plugin for Qoder IDE — shared skills, Qoder-specific enhancements (genui widgets, Canvas, Memory, CodeReview subagent, browser-use MCP). No hooks.json (Qoder uses skill-embedded gates instead of file-system event hooks).
-
-## STRUCTURE
-```
-plugins/qoder/
-├── README.md                      # Installation and usage guide
-├── AGENTS.md                      # This knowledge base
-├── skills/                        # Auto-populated by build-plugin.sh (8 skills)
-│   ├── sprint-flow/SKILL.md       # 7-phase pipeline + Qoder Pre-Edit Gate + Agent mapping
-│   ├── delphi-review/SKILL.md     # Multi-expert consensus + Qoder multi-model adaptation
-│   ├── test-specification-alignment/SKILL.md  # 2-phase test-spec verification
-│   ├── ralph-loop/SKILL.md        # REQ-level iterative build + Memory integration
-│   ├── test-driven-development/SKILL.md       # TDD enforcement (zero modification)
-│   ├── improve-codebase-architecture/SKILL.md # Architecture health + Canvas output
-│   ├── to-issues/SKILL.md         # Vertical slice issue splitting (zero modification)
-│   └── admin-template-guidelines/SKILL.md     # 6 maintainability rules (zero modification)
-├── widgets/                       # genui show_widget templates
-│   ├── quality-report.html        # Quality gate report dashboard
-│   └── sprint-dashboard.html      # Sprint status board
-└── references/
-    └── qoder-adaptation.md        # Platform differences and adaptation rules
-```
-
-## WHERE TO LOOK
-| Task | Location | Notes |
-|------|----------|-------|
-| Installation | README.md | User-level (~/.qoder/skills/) and project-level (.qoder/skills/) |
-| Skill adaptations | skills/*/SKILL.md | Qoder-specific paragraphs marked with "Qoder" header |
-| Pre-Edit Gate | skills/sprint-flow/SKILL.md | Replaces Claude Code's delphi-review-guard.sh hook |
-| Multi-model review | skills/delphi-review/references/qoder-multi-model.md | Expert→model mapping via Qoder Agent subagents |
-| Agent dispatch | skills/sprint-flow/SKILL.md | Phase→Qoder Agent mapping table |
-| External skill fallback | skills/sprint-flow/references/qoder-adaptation.md | superpowers/gstack replacements |
-| Quality widget | widgets/quality-report.html | genui show_widget template for Phase 3 results |
-| Sprint widget | widgets/sprint-dashboard.html | genui show_widget template for sprint status |
-| Memory integration | skills/ralph-loop/SKILL.md | UpdateMemory for permanent/contextual learnings |
-| CodeReview | skills/sprint-flow/SKILL.md | Phase 2 step 4 + Phase 3 code walkthrough |
-| Browser MCP | skills/sprint-flow/references/qoder-adaptation.md | browser-use replaces gstack/browse |
-
-## CONVENTIONS
-- Plugin skills auto-populated by build-plugin.sh --platform qoder — never edit plugins/qoder/skills/ directly
-- Qoder-specific skill modifications live in source skills/ directory (marked with "## Qoder" sections)
-- No hooks.json — Qoder lacks file-system event hooks; gates embedded in skill instructions
-- genui widgets use show_widget MCP tool with widget_path template mode
-- Memory system replaces gstack/learn for learnings persistence
-- CodeReview subagent replaces superpowers/requesting-code-review
-- browser-use MCP replaces gstack/browse for Phase 3 browser testing
-- External skills (superpowers/gstack) replaced by orchestrator inline execution or Qoder native capabilities
-
-## ANTI-PATTERNS (THIS PLUGIN)
-- Do NOT edit plugins/qoder/skills/ directly — rebuild from source via build-plugin.sh
-- Do NOT assume xp-gate CLI is installed — skills must degrade gracefully
-- Do NOT create hooks.json for Qoder — platform does not support file-system event hooks
-- Do NOT use curl for API calls — curl is in Qoder's command deny list; use PowerShell Invoke-RestMethod or Qoder's built-in multi-model capability
-- Do NOT reference OpenCode task() API in Qoder skills — use Agent tool (Browser/CodeReview/plan-agent) instead
-
-## UNIQUE STYLES
-- Qoder: SKILL.md + genui widgets + Canvas + Memory + CodeReview subagent + browser-use MCP
-- Pre-Edit Gate: skill-embedded mandatory check replaces Claude Code's physical hook blocking
-- Multi-model review: Qoder Agent subagent dispatch (plan-agent/CodeReview) replaces OpenCode task() API
-- Graceful degradation: skills work even without xp-gate CLI installed
-- Shared skill source: one SKILL.md → copied to Claude Code, OpenCode, and Qoder platforms
-
-## COMMANDS
-```bash
-# Build Qoder plugin
-npm run build:qoder-plugin                    # Build Qoder plugin with all 8 skills
-bash scripts/build-plugin.sh --platform qoder  # Same, via script
-
-# Install skills
-bash scripts/install-qoder-skills.sh --global  # User-level: ~/.qoder/skills/
-bash scripts/install-qoder-skills.sh --local   # Project-level: .qoder/skills/
-
-# Test plugin
-bash scripts/test-plugins.sh                   # Includes Qoder platform tests
-
-# CLI skill install
-xp-gate install-skill sprint-flow --platform qoder --global
-xp-gate install-skill delphi-review --platform qoder --local
-```
-
-## NOTES
-- v0.6.0+: Qoder plugin support introduced
-- Qoder plugin: 8 skills (all source skills, including admin-template-guidelines)
-- Claude Code plugin: 7 skills (no admin-template-guidelines)
-- OpenCode plugin: 7 skills + 3 custom tools (gate-check, gate-principles, gate-arch)
-- build-plugin.sh validates 8 expected skills for Qoder platform
-- test-plugins.sh: includes Qoder-specific test cases (widgets, no hooks.json)

package/plugins/qoder/README.md

@@ -1,87 +0,0 @@
-# XP-Gate Qoder Plugin
-
-Qoder IDE plugin exposing xp-gate quality gates, AI workflow skills, and visual dashboards.
-
-## Skills (8 AI Workflow Skills)
-
-| Skill | Trigger | Description |
-|-------|---------|-------------|
-| sprint-flow | `/sprint-flow "需求"` | 7-phase pipeline: THINK→PLAN→BUILD→REVIEW→ACCEPT→FEEDBACK→SHIP |
-| delphi-review | `/delphi-review` | Multi-expert anonymous consensus review (design + code-walkthrough) |
-| test-specification-alignment | `/test-specification-alignment` | 2-phase test-spec verification |
-| ralph-loop | Via sprint-flow Phase 2 | REQ-level iterative build (saves 40-67% tokens) |
-| test-driven-development | `/test-driven-development` | TDD enforcement (RED→GREEN→REFACTOR) |
-| improve-codebase-architecture | `/improve-codebase-architecture` | Architecture health checks |
-| to-issues | `/to-issues` | Vertical slice issue splitting |
-| admin-template-guidelines | Via sprint-flow BUILD | 6 maintainability rules for admin templates |
-
-## Qoder-Specific Enhancements
-
-- **genui Widgets**: Quality report dashboard and Sprint status board (via `show_widget`)
-- **CodeReview Subagent**: Deep integration for Phase 2 blind review and Phase 3 code walkthrough
-- **browser-use MCP**: Phase 3 browser automation testing (replaces gstack/browse)
-- **Memory System**: Learnings persistence via Qoder's built-in memory (replaces gstack/learn)
-- **Canvas Output**: Architecture relationship diagrams (via `.canvas.tsx`)
-
-## Installation
-
-### Option 1: User-Level (Global, All Projects)
-
-```bash
-# Via xp-gate CLI
-xp-gate install-skill sprint-flow --platform qoder --global
-xp-gate install-skill delphi-review --platform qoder --global
-
-# Or via install script (all 8 skills)
-bash scripts/install-qoder-skills.sh --global
-# Installs to: ~/.qoder/skills/
-```
-
-### Option 2: Project-Level (Current Project Only)
-
-```bash
-# Via install script
-bash scripts/install-qoder-skills.sh --local
-# Installs to: .qoder/skills/
-
-# Or via build script
-bash scripts/build-plugin.sh --platform qoder
-```
-
-### Option 3: Build from Source
-
-```bash
-npm run build:qoder-plugin
-# Produces plugins/qoder/ with all skills + widgets
-```
-
-## Requirements
-
-- Qoder IDE
-- xp-gate npm package installed globally (for git hooks: `npm install -g @boyingliu01/xp-gate && xp-gate init`)
-- Node.js ≥ 18 (for principles checker)
-
-## Pre-Edit Gate (Replaces Claude Code Hooks)
-
-Qoder does not have file-system event hooks like Claude Code's `PreToolUse`/`PostToolUse`. Instead, xp-gate skills embed **mandatory pre-edit checks** in their SKILL.md instructions:
-
-- **Delphi Gate**: Before any code edit in Phase 2+, skill verifies `.sprint-state/delphi-reviewed.json` verdict is `APPROVED`
-- **Principles Check**: After each REQ completion, skill runs principles checker via terminal
-
-## Graceful Degradation
-
-If xp-gate CLI is unavailable, skills provide helpful install instructions instead of failing. Quality gates still run via git hooks (pre-commit) at commit time.
-
-## Comparison with Other Platforms
-
-| Feature | Claude Code | OpenCode | Qoder |
-|---------|:-----------:|:--------:|:-----:|
-| AI Skills (8) | ✅ | ✅ | ✅ |
-| Git Hooks | ❌ (separate npm) | ❌ (separate npm) | ❌ (separate npm) |
-| Event Hooks | ✅ PreToolUse/PostToolUse | ✅ tool() | ❌ (skill-embedded) |
-| Custom Tools | ❌ | ✅ 3 tools | ❌ (MCP instead) |
-| Widget UI | ❌ | ❌ | ✅ genui |
-| Canvas Output | ❌ | ❌ | ✅ .canvas.tsx |
-| Memory System | ❌ | ❌ | ✅ UpdateMemory |
-| CodeReview Agent | ❌ | ❌ | ✅ subagent |
-| Browser MCP | ❌ | ❌ | ✅ browser-use |