@boyingliu01/xp-gate 0.35.17 → 0.35.18

package/plugins/claude-code/skills/sprint-flow/references/phase-5-ship.md

@@ -99,6 +99,20 @@ Sprint 分支已准备就绪。请选择操作：
 
 ### Step 2: 原生 Ship 步骤序列（选择 Option 2 时执行）
 
+#### Step 2.0: 完成定义后半段（DoD 第 7–10 项，MANDATORY）
+
+推送前逐项过 [`definition-of-done.md`](./definition-of-done.md)：
+
+| # | 条目 | 在这里怎么查 |
+|---|------|--------------|
+| 7 | 不做未经要求的相邻改动 | `git diff` 逐行对照 REQ；顺手发现的无关问题只提及，不删改 |
+| 8 | **已知限制已写明** | 没做的部分与原因写进提交说明或 issue，而不是留空 |
+| 9 | 文档与实现一致 | 门禁数量/CLI 表/阈值等出现在多处的事实同步更新（`AGENTS.md` "Known Drift"） |
+| 10 | 可回滚 | 提交粒度单一，能单独 revert 而不牵连其他改动 |
+
+第 8 项最容易被跳过，也最容易造成误导：**一个没写下来的限制，下一个人会按"已完成"来依赖它。**
+
+
 ```
 1. 运行测试:
    npm test  # 或项目特定测试命令

package/plugins/claude-code/skills/sprint-flow/scripts/verify-sprint-state.sh

@@ -3,6 +3,12 @@
 # Checks that sprint-state.json is valid JSON and all expected keys exist.
 #
 # Usage: bash verify-sprint-state.sh [sprint-state-dir]
+#
+# The expected keys are the ones `xp-gate sprint-init` actually writes (#679).
+# They were previously ['sprint_id','status','current_phase','phases'] -- a schema
+# this project has never written, so the check failed on every sprint the CLI
+# itself had just created. A verifier that rejects its own producer's output is
+# worse than no verifier: it trains people to ignore it.
 
 set -euo pipefail
 
@@ -24,12 +30,19 @@ node -e "
 const fs = require('fs');
 try {
     const data = JSON.parse(fs.readFileSync('$STATE_FILE', 'utf8'));
-    const requiredKeys = ['sprint_id', 'status', 'current_phase', 'phases'];
+    // These are the keys src/npm-package/lib/sprint-init.js writes.
+    const requiredKeys = ['id', 'task_description', 'phase', 'status'];
     const missing = requiredKeys.filter(k => !(k in data));
     if (missing.length > 0) {
         console.error('ERROR: Missing required keys in sprint-state.json:', missing.join(', '));
         process.exit(1);
     }
+    // phase_history is an array by construction; a non-array here means the file
+    // was hand-edited or written by a different tool, which is worth failing on.
+    if (!Array.isArray(data.phase_history)) {
+        console.error('ERROR: phase_history must be an array in sprint-state.json');
+        process.exit(1);
+    }
     console.log('OK: sprint-state.json is valid with all required keys');
 } catch (e) {
     console.error('ERROR: sprint-state.json is not valid JSON:', e.message);

package/plugins/claude-code/skills/test-specification-alignment/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/TEST-SPECIFICATION-ALIGNMENT KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 Test-Specification Alignment Engine — two-stage validation ensuring tests accurately reflect requirements and design specs.

package/plugins/dsh/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@boyingliu01/dsh-plugin-xp-gate",
-  "version": "0.35.17",
+  "version": "0.35.18",
   "description": "XP-Gate deterministic quality gates + AI workflow skills as a DeepSeek Harness (DSH) native plugin: gate-check / gate-principles / gate-arch tools + 12 SKILL.md skills",
   "type": "module",
   "main": "./lib/index.js",

package/plugins/dsh/skills/delphi-review/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/DELPHI-REVIEW KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 Delphi Consensus Review — multi-round anonymous expert review (≥90% threshold, exactly 3 experts with distinct executable model IDs). Supports design, requirements, and code-walkthrough modes.

package/plugins/dsh/skills/sprint-flow/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/SPRINT-FLOW KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 **6-phase** development pipeline (v2.0 compact redesign, Issue #290): PREP → DESIGN → BUILD → VERIFY → SHIP → CLOSE. Phase 3/6 BUILD default build mode is **ralph-loop** (REQ-level iteration, 40-67% token savings vs parallel). HARD-GATE between DESIGN (2/6) and BUILD (3/6): design must pass Delphi review (≥90% consensus) before any coding.

package/plugins/dsh/skills/sprint-flow/__tests__/verify-sprint-state.test.bats

@@ -0,0 +1,48 @@
+#!/usr/bin/env bats
+# @test REQ-688
+# @intent 验证 sprint-flow 的校验脚本能接受 `xp-gate sprint-init` 自己写出的状态文件
+#         ——脚本曾要求 sprint_id/current_phase/phases 这套从未被写入过的键，
+#         对 CLI 刚创建的 sprint 恒判失败（#679）
+# @covers AC-688-01, AC-688-02
+#
+# 为什么用真 sprint-init 而不是手写一个 fixture：缺陷的本质是「校验方与生产方对
+# schema 的认知不一致」。手写 fixture 会把「我认为 CLI 写什么」固化下来，双方一起
+# 错就一起绿。只有让生产方真的产出、校验方真的读，才能发现这类偏差。
+
+setup() {
+  REPO_ROOT="$(cd "$BATS_TEST_DIRNAME/../../.." && pwd)"
+  VERIFY="$REPO_ROOT/skills/sprint-flow/scripts/verify-sprint-state.sh"
+  CLI="$REPO_ROOT/src/npm-package/bin/xp-gate.js"
+  WORK="$(mktemp -d)"
+}
+
+teardown() {
+  rm -rf "$WORK"
+}
+
+@test "AC-688-01: accepts state written by xp-gate sprint-init" {
+  cd "$WORK"
+  git init -q .
+  echo '{}' > package.json
+  git add -A
+  git -c user.email=t@t -c user.name=t commit -q -m init
+
+  run node "$CLI" sprint-init "probe task"
+  [ -f ".sprint-state/sprint-state.json" ]
+
+  run bash "$VERIFY"
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"OK:"* ]]
+}
+
+@test "AC-688-02: still rejects a state file missing real keys" {
+  # The fix must not become "accept anything": a file that lost its task or its
+  # phase is genuinely broken and the verifier should say so.
+  mkdir -p "$WORK/.sprint-state"
+  printf '{"id":"s-1","phase":"BUILD"}\n' > "$WORK/.sprint-state/sprint-state.json"
+
+  cd "$WORK"
+  run bash "$VERIFY"
+  [ "$status" -eq 1 ]
+  [[ "$output" == *"task_description"* ]]
+}

package/plugins/dsh/skills/sprint-flow/references/definition-of-done.md

@@ -0,0 +1,47 @@
+# Definition of Done（统一完成定义）
+
+一个 REQ 只有在**全部 10 项**都可指向证据时才算完成。这份清单是 #607 的落地，
+对应指南 §12；它存在的理由是本仓库历史上反复出现「以为完成了，其实没有」：
+
+- **#561** 覆盖率阈值形状写错（`Path` 而非 `Partial<Record>`），声明 80% 但整个门禁是空转
+- **#570** 覆盖率检查 fail-open，跑不起来时报告通过
+- **#585** Gate 5c 静默通过
+- **.xp-gate/modules 陈旧副本**（2026-10-19）劫持 Gate 6 数周，hook 报的警告数和独立扫描永远对不上
+
+共同点全部是：**某一步从未真正执行，而没有任何东西把这件事说出来。**
+所以这份清单里每一项的判据都是「证据在哪儿」，不是「做没做」。
+
+---
+
+## 10 项完成定义
+
+| # | 条目 | 判据（必须能指出位置） |
+|---|------|------------------------|
+| 1 | **需求可追溯** | `specification.yaml` 有 REQ + AC；测试文件带 `@covers AC-xxx`；`scripts/__tests__/ac-reference-guard.test.cjs` 通过 |
+| 2 | **测试先行** | 实现前存在失败的测试；或说明为何不适用（纯文档/配置） |
+| 3 | **测试真的能失败** | 至少一个变异体/破坏被测试抓住。「测试通过」在从未运行时和「功能正确」无法区分 |
+| 4 | **验证命令已运行** | 记录实际命令与输出，不是「应该是绿的」。CI 命令与本地命令一致 |
+| 5 | **门禁未降级** | 无 `--no-verify`；无新增基线额度来吸收本次改动自身的告警 |
+| 6 | **镜像已同步** | 改了 `githooks/`/`skills/`/`plugins/` → `bash scripts/check-hook-mirror.sh` 输出 PASS |
+| 7 | **不做未经要求的相邻改动** | 每行改动可追溯到本 REQ；顺手发现的无关死代码只提及不删除 |
+| 8 | **已知限制已写明** | 没做的部分、以及为什么没做，写进 issue/提交，而不是留空 |
+| 9 | **文档与实现一致** | 门禁数量、CLI 表、阈值等出现在多处的事实同步更新（见 `AGENTS.md` "Known Drift"） |
+| 10 | **可回滚** | 提交粒度单一；出问题时能单独 revert 而不牵连其他改动 |
+
+---
+
+## 相位落点
+
+- **Phase 4 VERIFY**：逐项过 1/2/3/4/5/6，任何一项拿不出证据 → 不得进入 SHIP。
+- **Phase 5 SHIP**：逐项过 7/8/9/10，然后才推送。
+- **Phase 6 CLOSE**：把第 8 项「已知限制」回填到 issue 并关闭。
+
+## 常见误判
+
+| 看起来完成了 | 实际没有 |
+|---|---|
+| 「测试通过」 | 没跑过会失败的版本，不知道它能不能失败 |
+| 「门禁有配置」 | 配置形状错/键名不对时被静默丢弃，等于没有（#561） |
+| 「工具缺失时跳过」 | 跳过被当成通过；只有 Gate 8 是 fail-closed |
+| 「CI 绿了」 | CI 可能没跑到那一步（`if:` 条件不满足），或跑的是旧副本（#451） |
+| 「注释写了这么做」 | 注释与代码漂移；只有测试能固定行为 |

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

@@ -142,6 +142,22 @@ ELSE:
 - Web 前端附加: ui-review report + OPTIONAL QA/design-review/benchmark reports
 - 验证通过的 MVP
 
+### 完成定义（DoD 第 1–6 项）
+
+离开 VERIFY 前逐项过一遍 [`definition-of-done.md`](./definition-of-done.md)。**任何一项拿不出证据 → 不得进入 SHIP。**
+
+| # | 条目 | 在 VERIFY 的检查方式 |
+|---|------|----------------------|
+| 1 | 需求可追溯 | `npx vitest run scripts/__tests__/ac-reference-guard.test.cjs` 通过（测试引用的 AC 都在 spec 里） |
+| 2 | 测试先行 | 有 RED 证据（提交历史或记录），或说明为何不适用 |
+| 3 | **测试真的能失败** | 至少证明过一个变异体会被抓住——「测试通过」在从未运行时与「功能正确」无法区分 |
+| 4 | 验证命令已运行 | 记录实际命令 + 输出，不是「应该是绿的」 |
+| 5 | 门禁未降级 | 无 `--no-verify`；无为了本次改动新增的基线额度 |
+| 6 | 镜像已同步 | `bash scripts/check-hook-mirror.sh` → PASS |
+
+第 5 项值得单独说明：**加基线额度来吸收自己刚产生的告警，等于把门禁关掉一次。**
+`.warnings-baseline.json` 只应记录「改动之前就存在」的告警。
+
 ---
 
 ## Part B: FEEDBACK CAPTURE（反馈获取）

package/plugins/dsh/skills/sprint-flow/references/phase-5-ship.md

@@ -99,6 +99,20 @@ Sprint 分支已准备就绪。请选择操作：
 
 ### Step 2: 原生 Ship 步骤序列（选择 Option 2 时执行）
 
+#### Step 2.0: 完成定义后半段（DoD 第 7–10 项，MANDATORY）
+
+推送前逐项过 [`definition-of-done.md`](./definition-of-done.md)：
+
+| # | 条目 | 在这里怎么查 |
+|---|------|--------------|
+| 7 | 不做未经要求的相邻改动 | `git diff` 逐行对照 REQ；顺手发现的无关问题只提及，不删改 |
+| 8 | **已知限制已写明** | 没做的部分与原因写进提交说明或 issue，而不是留空 |
+| 9 | 文档与实现一致 | 门禁数量/CLI 表/阈值等出现在多处的事实同步更新（`AGENTS.md` "Known Drift"） |
+| 10 | 可回滚 | 提交粒度单一，能单独 revert 而不牵连其他改动 |
+
+第 8 项最容易被跳过，也最容易造成误导：**一个没写下来的限制，下一个人会按"已完成"来依赖它。**
+
+
 ```
 1. 运行测试:
    npm test  # 或项目特定测试命令

package/plugins/dsh/skills/sprint-flow/scripts/verify-sprint-state.sh

@@ -3,6 +3,12 @@
 # Checks that sprint-state.json is valid JSON and all expected keys exist.
 #
 # Usage: bash verify-sprint-state.sh [sprint-state-dir]
+#
+# The expected keys are the ones `xp-gate sprint-init` actually writes (#679).
+# They were previously ['sprint_id','status','current_phase','phases'] -- a schema
+# this project has never written, so the check failed on every sprint the CLI
+# itself had just created. A verifier that rejects its own producer's output is
+# worse than no verifier: it trains people to ignore it.
 
 set -euo pipefail
 
@@ -24,12 +30,19 @@ node -e "
 const fs = require('fs');
 try {
     const data = JSON.parse(fs.readFileSync('$STATE_FILE', 'utf8'));
-    const requiredKeys = ['sprint_id', 'status', 'current_phase', 'phases'];
+    // These are the keys src/npm-package/lib/sprint-init.js writes.
+    const requiredKeys = ['id', 'task_description', 'phase', 'status'];
     const missing = requiredKeys.filter(k => !(k in data));
     if (missing.length > 0) {
         console.error('ERROR: Missing required keys in sprint-state.json:', missing.join(', '));
         process.exit(1);
     }
+    // phase_history is an array by construction; a non-array here means the file
+    // was hand-edited or written by a different tool, which is worth failing on.
+    if (!Array.isArray(data.phase_history)) {
+        console.error('ERROR: phase_history must be an array in sprint-state.json');
+        process.exit(1);
+    }
     console.log('OK: sprint-state.json is valid with all required keys');
 } catch (e) {
     console.error('ERROR: sprint-state.json is not valid JSON:', e.message);

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

@@ -1,9 +1,9 @@
 # SKILLS/TEST-SPECIFICATION-ALIGNMENT KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 Test-Specification Alignment Engine — two-stage validation ensuring tests accurately reflect requirements and design specs.

package/plugins/dsh/src/delphi-findings.test.ts

@@ -39,13 +39,39 @@ function verdict(id: string, role: string, report: string, verdictValue: string,
   }
 }
 
+/**
+ * Build a ReviewOutcome from per-round verdicts.
+ *
+ * `consensusRatio` / `reachedConsensus` are computed the same way `runRound`
+ * does rather than stubbed: `RoundResult` declares them as required, and a
+ * fixture that hard-codes 0/false would be asserting against values the real
+ * code can never produce for these inputs. `formatOutcome` does not read them
+ * today (it takes consensus from the top-level fields), but a fixture that
+ * disagrees with the producer is exactly how a later change gets "verified"
+ * against something impossible.
+ *
+ * Delphi's seat/threshold constants are module-private in delphi.ts; they are
+ * restated here on purpose so this fixture does not force the producer to widen
+ * its public API just to be testable.
+ */
+const DELPHI_EXPERTS = 3
+const DELPHI_THRESHOLD_PERCENT = 90
+
 function outcomeOf(rounds: Array<{ round: number; verdicts: ReturnType<typeof verdict>[] }>): ReviewOutcome {
-  const mapped = rounds.map((r) => ({
-    round: r.round,
-    approvals: r.verdicts.filter((v) => v.verdict === "APPROVED").length,
-    validCount: r.verdicts.length,
-    verdicts: r.verdicts,
-  }))
+  const mapped = rounds.map((r) => {
+    const counted = r.verdicts.filter((v) => v.counted)
+    const approvals = counted.filter((v) => v.verdict === "APPROVED").length
+    // Mirrors runRound: consensus is measured over VALID responses only.
+    const consensusRatio = counted.length === 0 ? 0 : approvals / counted.length
+    return {
+      round: r.round,
+      approvals,
+      validCount: counted.length,
+      verdicts: r.verdicts,
+      consensusRatio,
+      reachedConsensus: counted.length === DELPHI_EXPERTS && consensusRatio * 100 >= DELPHI_THRESHOLD_PERCENT,
+    }
+  })
   return {
     rounds: mapped,
     consensus: false,

package/plugins/dsh/src/exit-code.test.ts

@@ -19,6 +19,7 @@
  * @covers AC-669-01, AC-669-02, AC-669-03, AC-669-04
  */
 import { execFileSync } from "node:child_process"
+import { describe, expect, it } from "vitest"
 import { buildCommand, detectDialectForShell } from "./command.js"
 import { classifyGateRun } from "./gate-runner.js"
 
@@ -68,6 +69,11 @@ describe("#669 命令串的退出码必须真实传递", () => {
     // 两者不一致时不能静默采信其中之一。
     const base = {
       timedOut: false,
+      // ShellRunResult requires both liveness fields. These fixtures represent a
+      // run that ended by itself, so "no signal" and "not aborted" are the honest
+      // values — omitting them only typechecked because this file never ran.
+      signal: null,
+      aborted: false,
       timeoutMs: 1000,
       stdout: { text: "some output\n[XP-Gate] exit:7\n", truncated: false },
       stderr: { text: "", truncated: false },

package/plugins/opencode/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@boyingliu01/opencode-plugin",
-  "version": "0.35.17",
+  "version": "0.35.18",
   "type": "module",
   "main": "index.ts",
   "description": "XP-Gate quality gates + AI workflow skills for OpenCode",

package/plugins/opencode/skills/delphi-review/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/DELPHI-REVIEW KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 Delphi Consensus Review — multi-round anonymous expert review (≥90% threshold, exactly 3 experts with distinct executable model IDs). Supports design, requirements, and code-walkthrough modes.

package/plugins/opencode/skills/sprint-flow/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/SPRINT-FLOW KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 **6-phase** development pipeline (v2.0 compact redesign, Issue #290): PREP → DESIGN → BUILD → VERIFY → SHIP → CLOSE. Phase 3/6 BUILD default build mode is **ralph-loop** (REQ-level iteration, 40-67% token savings vs parallel). HARD-GATE between DESIGN (2/6) and BUILD (3/6): design must pass Delphi review (≥90% consensus) before any coding.

package/plugins/opencode/skills/sprint-flow/__tests__/verify-sprint-state.test.bats

@@ -0,0 +1,48 @@
+#!/usr/bin/env bats
+# @test REQ-688
+# @intent 验证 sprint-flow 的校验脚本能接受 `xp-gate sprint-init` 自己写出的状态文件
+#         ——脚本曾要求 sprint_id/current_phase/phases 这套从未被写入过的键，
+#         对 CLI 刚创建的 sprint 恒判失败（#679）
+# @covers AC-688-01, AC-688-02
+#
+# 为什么用真 sprint-init 而不是手写一个 fixture：缺陷的本质是「校验方与生产方对
+# schema 的认知不一致」。手写 fixture 会把「我认为 CLI 写什么」固化下来，双方一起
+# 错就一起绿。只有让生产方真的产出、校验方真的读，才能发现这类偏差。
+
+setup() {
+  REPO_ROOT="$(cd "$BATS_TEST_DIRNAME/../../.." && pwd)"
+  VERIFY="$REPO_ROOT/skills/sprint-flow/scripts/verify-sprint-state.sh"
+  CLI="$REPO_ROOT/src/npm-package/bin/xp-gate.js"
+  WORK="$(mktemp -d)"
+}
+
+teardown() {
+  rm -rf "$WORK"
+}
+
+@test "AC-688-01: accepts state written by xp-gate sprint-init" {
+  cd "$WORK"
+  git init -q .
+  echo '{}' > package.json
+  git add -A
+  git -c user.email=t@t -c user.name=t commit -q -m init
+
+  run node "$CLI" sprint-init "probe task"
+  [ -f ".sprint-state/sprint-state.json" ]
+
+  run bash "$VERIFY"
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"OK:"* ]]
+}
+
+@test "AC-688-02: still rejects a state file missing real keys" {
+  # The fix must not become "accept anything": a file that lost its task or its
+  # phase is genuinely broken and the verifier should say so.
+  mkdir -p "$WORK/.sprint-state"
+  printf '{"id":"s-1","phase":"BUILD"}\n' > "$WORK/.sprint-state/sprint-state.json"
+
+  cd "$WORK"
+  run bash "$VERIFY"
+  [ "$status" -eq 1 ]
+  [[ "$output" == *"task_description"* ]]
+}

package/plugins/opencode/skills/sprint-flow/references/definition-of-done.md

@@ -0,0 +1,47 @@
+# Definition of Done（统一完成定义）
+
+一个 REQ 只有在**全部 10 项**都可指向证据时才算完成。这份清单是 #607 的落地，
+对应指南 §12；它存在的理由是本仓库历史上反复出现「以为完成了，其实没有」：
+
+- **#561** 覆盖率阈值形状写错（`Path` 而非 `Partial<Record>`），声明 80% 但整个门禁是空转
+- **#570** 覆盖率检查 fail-open，跑不起来时报告通过
+- **#585** Gate 5c 静默通过
+- **.xp-gate/modules 陈旧副本**（2026-10-19）劫持 Gate 6 数周，hook 报的警告数和独立扫描永远对不上
+
+共同点全部是：**某一步从未真正执行，而没有任何东西把这件事说出来。**
+所以这份清单里每一项的判据都是「证据在哪儿」，不是「做没做」。
+
+---
+
+## 10 项完成定义
+
+| # | 条目 | 判据（必须能指出位置） |
+|---|------|------------------------|
+| 1 | **需求可追溯** | `specification.yaml` 有 REQ + AC；测试文件带 `@covers AC-xxx`；`scripts/__tests__/ac-reference-guard.test.cjs` 通过 |
+| 2 | **测试先行** | 实现前存在失败的测试；或说明为何不适用（纯文档/配置） |
+| 3 | **测试真的能失败** | 至少一个变异体/破坏被测试抓住。「测试通过」在从未运行时和「功能正确」无法区分 |
+| 4 | **验证命令已运行** | 记录实际命令与输出，不是「应该是绿的」。CI 命令与本地命令一致 |
+| 5 | **门禁未降级** | 无 `--no-verify`；无新增基线额度来吸收本次改动自身的告警 |
+| 6 | **镜像已同步** | 改了 `githooks/`/`skills/`/`plugins/` → `bash scripts/check-hook-mirror.sh` 输出 PASS |
+| 7 | **不做未经要求的相邻改动** | 每行改动可追溯到本 REQ；顺手发现的无关死代码只提及不删除 |
+| 8 | **已知限制已写明** | 没做的部分、以及为什么没做，写进 issue/提交，而不是留空 |
+| 9 | **文档与实现一致** | 门禁数量、CLI 表、阈值等出现在多处的事实同步更新（见 `AGENTS.md` "Known Drift"） |
+| 10 | **可回滚** | 提交粒度单一；出问题时能单独 revert 而不牵连其他改动 |
+
+---
+
+## 相位落点
+
+- **Phase 4 VERIFY**：逐项过 1/2/3/4/5/6，任何一项拿不出证据 → 不得进入 SHIP。
+- **Phase 5 SHIP**：逐项过 7/8/9/10，然后才推送。
+- **Phase 6 CLOSE**：把第 8 项「已知限制」回填到 issue 并关闭。
+
+## 常见误判
+
+| 看起来完成了 | 实际没有 |
+|---|---|
+| 「测试通过」 | 没跑过会失败的版本，不知道它能不能失败 |
+| 「门禁有配置」 | 配置形状错/键名不对时被静默丢弃，等于没有（#561） |
+| 「工具缺失时跳过」 | 跳过被当成通过；只有 Gate 8 是 fail-closed |
+| 「CI 绿了」 | CI 可能没跑到那一步（`if:` 条件不满足），或跑的是旧副本（#451） |
+| 「注释写了这么做」 | 注释与代码漂移；只有测试能固定行为 |

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

@@ -142,6 +142,22 @@ ELSE:
 - Web 前端附加: ui-review report + OPTIONAL QA/design-review/benchmark reports
 - 验证通过的 MVP
 
+### 完成定义（DoD 第 1–6 项）
+
+离开 VERIFY 前逐项过一遍 [`definition-of-done.md`](./definition-of-done.md)。**任何一项拿不出证据 → 不得进入 SHIP。**
+
+| # | 条目 | 在 VERIFY 的检查方式 |
+|---|------|----------------------|
+| 1 | 需求可追溯 | `npx vitest run scripts/__tests__/ac-reference-guard.test.cjs` 通过（测试引用的 AC 都在 spec 里） |
+| 2 | 测试先行 | 有 RED 证据（提交历史或记录），或说明为何不适用 |
+| 3 | **测试真的能失败** | 至少证明过一个变异体会被抓住——「测试通过」在从未运行时与「功能正确」无法区分 |
+| 4 | 验证命令已运行 | 记录实际命令 + 输出，不是「应该是绿的」 |
+| 5 | 门禁未降级 | 无 `--no-verify`；无为了本次改动新增的基线额度 |
+| 6 | 镜像已同步 | `bash scripts/check-hook-mirror.sh` → PASS |
+
+第 5 项值得单独说明：**加基线额度来吸收自己刚产生的告警，等于把门禁关掉一次。**
+`.warnings-baseline.json` 只应记录「改动之前就存在」的告警。
+
 ---
 
 ## Part B: FEEDBACK CAPTURE（反馈获取）

package/plugins/opencode/skills/sprint-flow/references/phase-5-ship.md

@@ -99,6 +99,20 @@ Sprint 分支已准备就绪。请选择操作：
 
 ### Step 2: 原生 Ship 步骤序列（选择 Option 2 时执行）
 
+#### Step 2.0: 完成定义后半段（DoD 第 7–10 项，MANDATORY）
+
+推送前逐项过 [`definition-of-done.md`](./definition-of-done.md)：
+
+| # | 条目 | 在这里怎么查 |
+|---|------|--------------|
+| 7 | 不做未经要求的相邻改动 | `git diff` 逐行对照 REQ；顺手发现的无关问题只提及，不删改 |
+| 8 | **已知限制已写明** | 没做的部分与原因写进提交说明或 issue，而不是留空 |
+| 9 | 文档与实现一致 | 门禁数量/CLI 表/阈值等出现在多处的事实同步更新（`AGENTS.md` "Known Drift"） |
+| 10 | 可回滚 | 提交粒度单一，能单独 revert 而不牵连其他改动 |
+
+第 8 项最容易被跳过，也最容易造成误导：**一个没写下来的限制，下一个人会按"已完成"来依赖它。**
+
+
 ```
 1. 运行测试:
    npm test  # 或项目特定测试命令

package/plugins/opencode/skills/sprint-flow/scripts/verify-sprint-state.sh

@@ -3,6 +3,12 @@
 # Checks that sprint-state.json is valid JSON and all expected keys exist.
 #
 # Usage: bash verify-sprint-state.sh [sprint-state-dir]
+#
+# The expected keys are the ones `xp-gate sprint-init` actually writes (#679).
+# They were previously ['sprint_id','status','current_phase','phases'] -- a schema
+# this project has never written, so the check failed on every sprint the CLI
+# itself had just created. A verifier that rejects its own producer's output is
+# worse than no verifier: it trains people to ignore it.
 
 set -euo pipefail
 
@@ -24,12 +30,19 @@ node -e "
 const fs = require('fs');
 try {
     const data = JSON.parse(fs.readFileSync('$STATE_FILE', 'utf8'));
-    const requiredKeys = ['sprint_id', 'status', 'current_phase', 'phases'];
+    // These are the keys src/npm-package/lib/sprint-init.js writes.
+    const requiredKeys = ['id', 'task_description', 'phase', 'status'];
     const missing = requiredKeys.filter(k => !(k in data));
     if (missing.length > 0) {
         console.error('ERROR: Missing required keys in sprint-state.json:', missing.join(', '));
         process.exit(1);
     }
+    // phase_history is an array by construction; a non-array here means the file
+    // was hand-edited or written by a different tool, which is worth failing on.
+    if (!Array.isArray(data.phase_history)) {
+        console.error('ERROR: phase_history must be an array in sprint-state.json');
+        process.exit(1);
+    }
     console.log('OK: sprint-state.json is valid with all required keys');
 } catch (e) {
     console.error('ERROR: sprint-state.json is not valid JSON:', e.message);

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

@@ -1,9 +1,9 @@
 # SKILLS/TEST-SPECIFICATION-ALIGNMENT KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 Test-Specification Alignment Engine — two-stage validation ensuring tests accurately reflect requirements and design specs.

package/plugins/qoder/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "xp-gate",
-  "version": "0.35.17",
+  "version": "0.35.18",
   "displayName": "XP-Gate",
   "description": "Extreme Programming quality gates + AI workflow skills for Qoder. Includes 12 pre-commit gates (Gate 0-11), 8 pre-push gates, Sprint Flow (6 phases), and Delphi multi-expert review (>=90% consensus).",
   "author": {

package/plugins/qoder/skills/delphi-review/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/DELPHI-REVIEW KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 Delphi Consensus Review — multi-round anonymous expert review (≥90% threshold, exactly 3 experts with distinct executable model IDs). Supports design, requirements, and code-walkthrough modes.

package/plugins/qoder/skills/sprint-flow/AGENTS.md

@@ -1,9 +1,9 @@
 # SKILLS/SPRINT-FLOW KNOWLEDGE BASE
 
 **Generated:** 2026-10-31
-**Commit:** 26a741de
+**Commit:** 97e3c2ec
 **Branch:** main
-**Version:** 0.35.17.0
+**Version:** 0.35.18.0
 
 ## OVERVIEW
 **6-phase** development pipeline (v2.0 compact redesign, Issue #290): PREP → DESIGN → BUILD → VERIFY → SHIP → CLOSE. Phase 3/6 BUILD default build mode is **ralph-loop** (REQ-level iteration, 40-67% token savings vs parallel). HARD-GATE between DESIGN (2/6) and BUILD (3/6): design must pass Delphi review (≥90% consensus) before any coding.