@boyingliu01/xp-gate 0.18.5 → 0.19.0

package/skills/sprint-flow/references/phase-2-design.md

@@ -59,13 +59,15 @@ grill-with-docs 执行：
 grill 访谈达成共享理解后（或 CONTEXT.md 快速路径下读取已有上下文后）、设计文档生成前，调用需求评审：
 
 ```
-npx xp-gate delphi-review --mode requirements
+/delphi-review --mode requirements
 ```
 
-**评审配置**（轻量）：
+这是 Agent skill 调用，不是 `xp-gate` npm CLI 子命令。Skill orchestrator 必须分别调用 per-expert runner 三次，每次保留 `--expert <architecture|technical|feasibility> --mode requirements`，再聚合三份结果；禁止用一个 runner 进程执行全部模型。
+
+**评审配置**：
 - 复用现有 3 专家（architecture/feasibility/technical），`--mode requirements` 切换评审焦点提示词
-- 轻量配置：2 专家、1 轮
-- lightweight sprint（`change_type == "修改已存在代码"`）：跳过 R1，需求评审合并入 R2
+- 所有路径固定使用 3 专家，Round 1 独立执行并验证 distinct model IDs，最多 5 轮
+- force level 只调整上下文深度与迭代预算；所有路径都执行 R1，不减少专家数
 
 **评审焦点**：
 - 用户场景遗漏
@@ -81,8 +83,16 @@ npx xp-gate delphi-review --mode requirements
 {
   "verdict": "APPROVED | GAPS_FOUND",
   "timestamp": "<ISO 8601>",
-  "requirements_hash": "<SHA-256 of 需求陈述 + CONTEXT.md 内容>",
-  "experts": 2,
+  "requirements_statement": "<被评审的原始需求陈述>",
+  "context_file_used": "CONTEXT.md | null",
+  "requirements_hash": "<SHA-256 of requirements_statement + context file exact UTF-8 content if used + timestamp YYYY-MM-DD>",
+  "head_commit": "<git rev-parse HEAD>",
+  "consensus_ratio": 1.0,
+  "expert_verdicts": [
+    { "role": "architecture", "verdict": "APPROVED", "result_type": "delphi_expert_result", "requested_model": "provider/model-a" },
+    { "role": "technical", "verdict": "APPROVED", "result_type": "delphi_expert_result", "requested_model": "provider/model-b" },
+    { "role": "feasibility", "verdict": "APPROVED", "result_type": "delphi_expert_result", "requested_model": "provider/model-c" }
+  ],
   "rounds": 1,
   "gaps": []
 }
@@ -90,10 +100,10 @@ npx xp-gate delphi-review --mode requirements
 
 **GAPS_FOUND 处理**：
 - 回到 Step 1 补充访谈（或补充 CONTEXT.md 内容）
-- 最多 2 轮循环后升级给用户决策
+- 最多 5 轮循环后升级给用户决策
 
 **程序化校验**：
-- `phase-transition 2 completed` 校验该文件存在、`verdict=APPROVED` 且 `requirements_hash` 匹配当前需求内容
+- `phase-transition 2 completed` 校验 schema-v2 必填字段、`verdict=APPROVED`、三份专家证据、`consensus_ratio`、当前 HEAD，并从需求陈述、可选 context 文件和 timestamp 日期重新计算 `requirements_hash`
 - 不匹配 → BLOCK（防陈旧绑定）
 
 ### Step 3: 原生设计文档生成
@@ -125,7 +135,7 @@ orchestrator 基于访谈记录 + CONTEXT.md + R1 评审结论生成设计文档
 
 | 暂停点 | 触发条件 | 用户操作 | 自动恢复条件 |
 |--------|---------|---------|-------------|
-| R1 GAPS_FOUND | 需求评审发现缺口 | 补充访谈/上下文 | 重新 R1 评审（最多 2 轮） |
+| R1 GAPS_FOUND | 需求评审发现缺口 | 补充访谈/上下文 | 重新 R1 评审（最多 5 轮） |
 | **HARD-GATE** | 设计文档未 APPROVED | 用户审批设计文档 | 设计 APPROVED 后自动进入 Part B |
 
 ### 输出
@@ -149,7 +159,7 @@ orchestrator 基于访谈记录 + CONTEXT.md + R1 评审结论生成设计文档
 
 IF change_type == "修改已存在代码":
   → 增量优化路径: SKIP batch-grill-me
-  → 直接进入 Step 7: lightweight R2 delphi-review (2 专家, 1 轮)
+   → 直接进入 Step 7: R2 delphi-review（3 专家，最多 5 轮）
 ELSE (change_type == "新增功能" 或 未定义):
   → 标准路径: 继续 Step 6 (batch-grill-me) → Step 7
 ```
@@ -158,7 +168,7 @@ ELSE (change_type == "新增功能" 或 未定义):
 
 | change_type | batch-grill-me | R2 delphi-review |
 |------------|----------------|------------------|
-| `修改已存在代码` | ❌ SKIP | lightweight (2 专家, 1 轮) |
+| `修改已存在代码` | ❌ SKIP | 3 专家，最多 5 轮 |
 | `新增功能` | ✅ 执行 | 标准 (3 专家) |
 | `undefined` / 缺失 | ✅ 执行 | 标准 (3 专家) |
 
@@ -183,12 +193,12 @@ batch-grill-me 替代 autoplan 的 taste_decisions 功能：
 # 标准路径（3 专家）
 skill(name="delphi-review", user_message="[设计文档 + batch-grill-me 决策结果]")
 
-# 增量优化路径（2 专家, 1 轮 — 来自 Step 5 路由）
-skill(name="delphi-review", user_message="[设计文档]", experts=2, max_rounds=1)
+# 增量优化路径（3 专家，最多 5 轮）
+skill(name="delphi-review", user_message="[设计文档]", experts=3, max_rounds=5)
 ```
 
 - **标准路径**: Round 1: 3 专家匿名独立评审 → Round 2+: 交换意见直到共识
-- **增量优化路径**: 2 专家, 1 轮
+- **增量优化路径**: 3 专家，最多 5 轮
 - ≥90% 共识 + APPROVED 才通过
 - 输出: APPROVED / REQUEST_CHANGES
 

package/skills/sprint-flow/references/phase-4-verify.md

@@ -18,7 +18,7 @@ Web 前端项目额外增加：原生 UI 审查 + 可选增强。
 ### 调用 Skills
 
 **所有项目（原生 HARD-GATE 链）**:
-- `delphi-review --mode code-walkthrough` — 多专家匿名代码走查（2-3 domestic models, ≥90% consensus）
+- `delphi-review --mode code-walkthrough` — 三位专家匿名代码走查，验证三个 distinct requested_model 并聚合 ≥90% consensus
 - `test-specification-alignment` — 测试与 Spec 对齐验证（#367 程序化 HARD-GATE）
 - `xp-gate check --all` — 全量质量门禁（Gate 0–9，含安全审计）
 
@@ -50,11 +50,11 @@ ELSE → SKIP 并记录 "[sprint-flow] 浏览器验证 SKIP（无可用浏览器
 skill(name="delphi-review", user_message="--mode code-walkthrough")
 ```
 
-- 2-3 位国内模型专家匿名独立评审（DeepSeek-v4-pro + Kimi-K2.6 + Qwen3.6-Plus）
-- Round 1: 匿名独立评审（防止 anchoring bias）
-- Round 2: 交换意见，响应关切
-- Round 3: 最终立场（如需）
-- ≥90% 共识 + APPROVED 才通过
+- architecture、technical、feasibility 三位专家匿名独立评审，模型 ID 必须 distinct
+- Round 1: 三位专家独立执行，输出 `result_type=delphi_expert_result`
+- 验证三份成功结果和三个 distinct requested_model 后再聚合
+- 后续轮次最多到 Round 5，≥90% 共识 + APPROVED 才通过
+- 任一失败或模型重复都阻断，不降级为少数专家
 
 **如果 REQUEST_CHANGES**: 暂停等待用户处理 → 修复后重新评审。**如果 APPROVED**: 写入 `.code-walkthrough-result.json`（1 小时有效期）→ 进入 Step 2。
 

package/skills/sprint-flow/references/phase-overview.md

@@ -36,7 +36,7 @@ Sprint Flow is NOT just a sequential launcher of existing skills. Here's what ma
 
 1. **40-67% Token Savings via Ralph Loop**: Phase 3 BUILD default (`ralph-loop`) processes one REQ at a time with clean context, avoiding the linear context accumulation that costs 150K+ tokens in parallel mode.
 
-2. **HARD-GATE Discipline**: Design must pass Delphi review (≥90% consensus, ≥2 model providers, domestic models only) before Phase 3 BUILD can start. This is enforced both in SKILL.md instructions and via the Claude Code plugin's PreToolUse hook.
+2. **HARD-GATE Discipline**: Design must pass Delphi review with all three expert roles executed successfully, three distinct requested model IDs verified, and ≥90% aggregate consensus within five rounds before Phase 3 BUILD can start. This is enforced both in SKILL.md instructions and via the Claude Code plugin's PreToolUse hook.
 
 3. **Emergent Requirements Acknowledgment**: Based on research showing 78% of failures are invisible to AI (arXiv study), Phase 6 CLOSE includes mandatory manual UAT verification — cannot be automated, skipped, or bypassed.
 
@@ -155,7 +155,7 @@ Sprint Flow: PREP → DESIGN → BUILD → VERIFY → SHIP → CLOSE
 **快速参考**:
 - **Step 0: CONTEXT.md 预检 (#322)**: 存在则 SKIP grill 访谈，但 R1 需求评审仍执行（CONTEXT.md 可能陈旧）
 - **Step 1: grill-with-docs** — `skill(name="grill-with-docs")` — 逐个追问 + CONTEXT.md/ADR 沉淀。**HARD-GATE**: 设计未批准 → 不可进入实现
-- **Step 2: R1 需求评审 (#368)** — `npx xp-gate delphi-review --mode requirements` — 轻量 2 专家 1 轮，输出 requirements-reviewed.json（含 requirements_hash 防陈旧绑定）。lightweight sprint 跳过 R1 合并入 R2
+- **Step 2: R1 需求评审 (#368)** — `/delphi-review --mode requirements` Agent skill — 所有路径都由三专家独立执行，验证 distinct model IDs 后聚合 ≥90% 共识，输出 requirements-reviewed.json（含可重新计算的 requirements_hash 与 exact-HEAD 防陈旧绑定）
 - **Step 3: 原生设计文档生成** — `docs/plans/YYYY-MM-DD-<topic>-design.md`（需求摘要、2-3 方案、推荐、成功标准）
 - **Step 4: HARD-GATE APPROVAL** — 用户审批设计文档
 - **Step 5: 路由分叉** — `change_type == "修改已存在代码"` → SKIP batch-grill-me, lightweight R2

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

@@ -1,9 +1,9 @@
 # SKILLS/TEST-SPECIFICATION-ALIGNMENT KNOWLEDGE BASE
 
-**Generated:** 2026-07-29
-**Commit:** 05d23ae
-**Branch:** main
-**Version:** 0.18.5.0
+**Generated:** 2026-08-24
+**Commit:** a1a4683
+**Branch:** dsh-plugin
+**Version:** 0.19.0.0
 
 ## OVERVIEW
 Test-Specification Alignment Engine — two-stage validation ensuring tests accurately reflect requirements and design specs.

package/sprint-gate.sh

@@ -0,0 +1,291 @@
+#!/bin/bash
+# sprint-gate.sh — Sprint Flow Validation Gate
+#
+# PURPOSE: Enforce Sprint Flow phase discipline at git hook level.
+# Validates delphi-review approval, sprint state, and specification
+# artifacts before allowing commits and pushes on sprint branches.
+#
+# INTEGRATION: Called from pre-commit (Gate 10) and pre-push (Gate S),
+# or run independently for CI/manual validation.
+#
+# USAGE:
+#   sprint-gate.sh --pre-commit   # Validate before commit
+#   sprint-gate.sh --pre-push     # Validate before push
+#
+# MECHANISM:
+#   --pre-commit:
+#     - If .sprint-state/ missing → SKIP (not a sprint project)
+#     - If sprint-state.json missing/invalid → SKIP (non-sprint commit)
+#     - If phase >= 1 (PLAN completed) and delphi-reviewed.json missing → DENY
+#     - If delphi-reviewed.json exists, verdict must be "APPROVED"
+#     - If jq AND node unavailable → WARN but ALLOW (graceful degradation)
+#     - If jq unavailable but node available → use Node.js JSON parser fallback
+#
+#   --pre-push:
+#     - If .sprint-state/ missing → SKIP
+#     - If branch doesn't match sprint/* → SKIP
+#     - If delphi-reviewed.json missing or verdict != "APPROVED" → DENY
+#     - If phase < 2 (BUILD not started) → DENY
+#     - If phase >= 2 and specification.yaml missing → DENY
+#     - If jq AND node unavailable → DENY (cannot verify verdict)
+#
+# GRACEFUL DEGRADATION:
+#   - If .sprint-state/ directory doesn't exist → SKIP (not a sprint project)
+#   - If sprint-state.json is missing/invalid → SKIP (non-sprint commit)
+#   - JSON parser fallback chain: jq → Node.js → explicit degradation message
+#
+# OUTPUT: JSON to stdout — {"decision":"allow|deny|skip",...}
+# EXIT: 0 = pass/skip, 1 = block
+
+set -euo pipefail
+
+# --- Argument parsing ---
+MODE=""
+case "${1:-}" in
+  --pre-commit) MODE="pre-commit" ;;
+  --pre-push)   MODE="pre-push" ;;
+  *)
+    echo '{"decision":"deny","reason":"Usage: sprint-gate.sh --pre-commit | --pre-push"}'
+    exit 1
+    ;;
+esac
+
+# --- Locate repo root and sprint state ---
+REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null)" || {
+  echo '{"decision":"skip","reason":"not inside a git repository"}'
+  exit 0
+}
+
+SPRINT_STATE_DIR="$REPO_ROOT/.sprint-state"
+SPRINT_STATE_JSON="$SPRINT_STATE_DIR/sprint-state.json"
+APPROVED_FILE="$SPRINT_STATE_DIR/delphi-reviewed.json"
+
+# If no .sprint-state directory, this isn't a sprint project → SKIP
+if [ ! -d "$SPRINT_STATE_DIR" ]; then
+  echo '{"decision":"skip","reason":"not a sprint project"}'
+  exit 0
+fi
+
+# --- Helper: check tool availability ---
+HAS_JQ=false
+if command -v jq &>/dev/null; then
+  HAS_JQ=true
+fi
+
+HAS_NODE=false
+if command -v node &>/dev/null; then
+  HAS_NODE=true
+fi
+
+# JSON_PARSER: "jq" | "node" | "none"
+if $HAS_JQ; then
+  JSON_PARSER="jq"
+elif $HAS_NODE; then
+  JSON_PARSER="node"
+else
+  JSON_PARSER="none"
+fi
+
+# --- Helper: validate JSON file ---
+# Returns 0 if valid, 1 if invalid or no parser available
+json_validate() {
+  local file="$1"
+  if [ ! -f "$file" ]; then
+    return 1
+  fi
+  case "$JSON_PARSER" in
+    jq)
+      jq empty "$file" 2>/dev/null
+      return $?
+      ;;
+    node)
+      node -e "try{JSON.parse(require('fs').readFileSync('$file','utf8'));process.exit(0)}catch(e){process.exit(1)}" 2>/dev/null
+      return $?
+      ;;
+    *)
+      return 1
+      ;;
+  esac
+}
+
+# --- Helper: extract JSON field ---
+# Usage: json_extract <file> <jq-filter>
+# jq-filter is a jq expression like '.phase // -1' or '.verdict // ""'
+# Falls back to Node.js when jq is unavailable
+json_extract() {
+  local file="$1"
+  local filter="$2"
+  case "$JSON_PARSER" in
+    jq)
+      jq -r "$filter" "$file" 2>/dev/null
+      return $?
+      ;;
+    node)
+      # Translate simple jq filters to Node.js
+      # Supports: .field // default patterns
+      local field default_val
+      field=$(echo "$filter" | sed 's/^\.//;s/ *\/\/.*$//')
+      default_val=$(echo "$filter" | sed -n 's/.*\/\/ *//p')
+      node -e "
+        try {
+          var d=JSON.parse(require('fs').readFileSync('$file','utf8'));
+          var v=d['${field}'];
+          if(v===undefined||v===null){v=${default_val:-null}}
+          process.stdout.write(String(v));
+        }catch(e){process.stdout.write('${default_val:-}');process.exit(1)}
+      " 2>/dev/null
+      return $?
+      ;;
+    *)
+      echo ""
+      return 1
+      ;;
+  esac
+}
+
+# --- Helper: read phase from sprint-state.json ---
+# Returns phase number or -1 if unreadable
+read_phase() {
+  if [ ! -f "$SPRINT_STATE_JSON" ]; then
+    echo "-1"
+    return
+  fi
+  if [ "$JSON_PARSER" = "none" ]; then
+    echo "-1"
+    return
+  fi
+  if ! json_validate "$SPRINT_STATE_JSON"; then
+    echo "-1"
+    return
+  fi
+  local phase
+  phase=$(json_extract "$SPRINT_STATE_JSON" '.phase // -1') || phase="-1"
+  echo "$phase"
+}
+
+# --- Helper: read delphi verdict ---
+# Returns verdict string or empty if unreadable
+read_verdict() {
+  if [ ! -f "$APPROVED_FILE" ]; then
+    echo ""
+    return
+  fi
+  if [ "$JSON_PARSER" = "none" ]; then
+    echo ""
+    return
+  fi
+  if ! json_validate "$APPROVED_FILE"; then
+    echo ""
+    return
+  fi
+  json_extract "$APPROVED_FILE" '.verdict // ""' || echo ""
+}
+
+# --- Pre-commit mode ---
+if [ "$MODE" = "pre-commit" ]; then
+  PHASE=$(read_phase)
+
+  # If sprint-state.json is missing or invalid, skip (non-sprint commit)
+  if [ "$PHASE" = "-1" ]; then
+    if [ ! -f "$SPRINT_STATE_JSON" ]; then
+      echo '{"decision":"skip","reason":"sprint-state.json not found, non-sprint commit"}'
+    else
+      echo '{"decision":"skip","warning":"sprint-state.json is not valid JSON or no JSON parser available (install jq or node), allowing commit"}'
+    fi
+    exit 0
+  fi
+
+  # Phase >= 1 means PLAN is completed — delphi-review must be APPROVED
+  if [ "$PHASE" -ge 1 ] 2>/dev/null; then
+    # Check if delphi-reviewed.json exists
+    if [ ! -f "$APPROVED_FILE" ]; then
+      echo '{"decision":"deny","reason":"delphi-review not APPROVED. Complete Phase 1 before committing code changes."}'
+      exit 1
+    fi
+
+    # If no JSON parser available, warn but allow (graceful degradation)
+    if [ "$JSON_PARSER" = "none" ]; then
+      echo '{"decision":"allow","warning":"neither jq nor node available, cannot verify delphi-review verdict. Install jq or node for full protection."}'
+      exit 0
+    fi
+
+    # Validate delphi-reviewed.json is valid JSON
+    if ! json_validate "$APPROVED_FILE"; then
+      echo '{"decision":"deny","reason":"delphi-reviewed.json is not valid JSON. Re-run: /delphi-review"}'
+      exit 1
+    fi
+
+    # Check verdict
+    VERDICT=$(read_verdict)
+    if [ "$VERDICT" != "APPROVED" ]; then
+      echo "{\"decision\":\"deny\",\"reason\":\"delphi-review verdict is '${VERDICT}', not APPROVED. Fix issues and re-run: /delphi-review\"}"
+      exit 1
+    fi
+
+    # All checks passed
+    echo "{\"decision\":\"allow\",\"message\":\"delphi-review APPROVED, sprint phase ${PHASE} validated\"}"
+    exit 0
+  fi
+
+  # Phase < 1 (pre-PLAN) — no delphi enforcement needed
+  echo "{\"decision\":\"allow\",\"reason\":\"sprint phase ${PHASE} (pre-PLAN), no delphi enforcement required\"}"
+  exit 0
+fi
+
+# --- Pre-push mode ---
+if [ "$MODE" = "pre-push" ]; then
+  # Only enforce on sprint/* branches
+  BRANCH="$(git branch --show-current 2>/dev/null)" || BRANCH=""
+  case "$BRANCH" in
+    sprint/*) ;; # continue
+    *)
+      echo "{\"decision\":\"skip\",\"reason\":\"branch '${BRANCH}' does not match sprint/* pattern\"}"
+      exit 0
+      ;;
+  esac
+
+  # Validate sprint-state.json exists and is readable
+  PHASE=$(read_phase)
+  if [ "$PHASE" = "-1" ] && [ ! -f "$SPRINT_STATE_JSON" ]; then
+    echo "{\"decision\":\"deny\",\"reason\":\"sprint-state.json not found on sprint branch '${BRANCH}'\"}"
+    exit 1
+  fi
+
+  # Delphi review must be APPROVED
+  if [ ! -f "$APPROVED_FILE" ]; then
+    echo '{"decision":"deny","reason":"delphi-reviewed.json not found. Complete Phase 1 delphi-review before pushing."}'
+    exit 1
+  fi
+
+  if [ "$JSON_PARSER" = "none" ]; then
+    echo '{"decision":"deny","reason":"neither jq nor node available, cannot verify delphi-review verdict. Install jq or node to push from sprint branch."}'
+    exit 1
+  fi
+
+  if ! json_validate "$APPROVED_FILE"; then
+    echo '{"decision":"deny","reason":"delphi-reviewed.json is not valid JSON. Re-run: /delphi-review"}'
+    exit 1
+  fi
+
+  VERDICT=$(read_verdict)
+  if [ "$VERDICT" != "APPROVED" ]; then
+    echo "{\"decision\":\"deny\",\"reason\":\"delphi-review verdict is '${VERDICT}', not APPROVED. Fix issues and re-run: /delphi-review\"}"
+    exit 1
+  fi
+
+  # specification.yaml must exist (Phase 1 PLAN output)
+  SPEC_FILE="$REPO_ROOT/specification.yaml"
+  if [ ! -f "$SPEC_FILE" ]; then
+    # Also check sprint-state output directory
+    SPEC_FILE="$SPRINT_STATE_DIR/phase-outputs/specification.yaml"
+    if [ ! -f "$SPEC_FILE" ]; then
+      echo '{"decision":"deny","reason":"specification.yaml not found. Phase 1 PLAN must produce specification.yaml before pushing."}'
+      exit 1
+    fi
+  fi
+
+  # All checks passed — comprehensive success output
+  SPEC_SIZE=$(wc -c < "$SPEC_FILE" 2>/dev/null | tr -d ' ')
+  echo "{\"decision\":\"allow\",\"message\":\"sprint push validated\",\"branch\":\"${BRANCH}\",\"phase\":${PHASE},\"verdict\":\"APPROVED\",\"specification_size_bytes\":${SPEC_SIZE:-0}}"
+  exit 0
+fi