harness：迁到独立引擎 delivery-harness 0.1.0（T008）

.claude/settings.json

@@ -6,7 +6,7 @@
         "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$CLAUDE_PROJECT_DIR/harness/command_guard.py\" --format claude --role designer"
+            "command": "python3 \"$CLAUDE_PROJECT_DIR/.harness/engine/cli.py\" guard-command --format claude --role designer"
           }
         ]
       },
@@ -15,7 +15,7 @@
         "hooks": [
           {
             "type": "command",
-            "command": "python3 \"$CLAUDE_PROJECT_DIR/harness/command_guard.py\" --format claude --role designer"
+            "command": "python3 \"$CLAUDE_PROJECT_DIR/.harness/engine/cli.py\" guard-command --format claude --role designer"
           }
         ]
       }

.codex/hooks.json

@@ -5,7 +5,7 @@
         "hooks": [
           {
             "type": "command",
-            "command": "sh -c 'exec python3 \"$(git rev-parse --show-toplevel)/harness/command_guard.py\" --format claude --role designer'",
+            "command": "sh -c 'exec python3 \"$(git rev-parse --show-toplevel)/.harness/engine/cli.py\" guard-command --format claude --role designer'",
             "timeout": 30
           }
         ]

.githooks/_runtime

@@ -1,13 +1,17 @@
-# harness 钩子公共部分（由各钩子 source）：定位 harness 与解释器。
-# 解释器优先用开发包运行时 venv，与 bin/verify 一致；找不到时按 HARNESS_FAIL_OPEN 决定放行或拒绝。
+# delivery-harness 钩子公共部分（由各钩子 source）：定位引擎与解释器。
+# 解释器优先用 .harness/config/checks.toml [runtime] python（相对仓库根），与 bin/harness 一致；
+# 没有配置或不可执行时用 PATH 上的 python3；都找不到时按 HARNESS_FAIL_OPEN 决定放行或拒绝。
 HARNESS_ROOT=$(CDPATH= cd -- "$(dirname -- "$0")/.." && pwd)
-HARNESS_PY="$HARNESS_ROOT/scratch/iterm-probe-venv/bin/python"
-if [ ! -x "$HARNESS_PY" ]; then
+HARNESS_CLI="$HARNESS_ROOT/.harness/engine/cli.py"
+HARNESS_PY=$(sed -n '/^\[runtime\]/,/^\[/s/^python *= *"\(.*\)".*/\1/p' "$HARNESS_ROOT/.harness/config/checks.toml" 2>/dev/null | head -n 1)
+if [ -n "$HARNESS_PY" ]; then
+    HARNESS_PY="$HARNESS_ROOT/$HARNESS_PY"
+fi
+if [ -z "$HARNESS_PY" ] || [ ! -x "$HARNESS_PY" ]; then
     HARNESS_PY=$(command -v python3 || true)
 fi
 if [ -z "$HARNESS_PY" ]; then
     echo "harness：找不到 python3，$(basename -- "$0") 守卫未执行。" >&2
     if [ "${HARNESS_FAIL_OPEN:-0}" = 1 ]; then exit 0; fi
     exit 1
 fi
-HARNESS_GUARD="$HARNESS_ROOT/harness/git_guard.py"

.githooks/pre-commit

@@ -1,4 +1,4 @@
 #!/bin/sh
-# 保护分支上禁止提交；暂存区卫生；快速 verify（harness/git_guard.py）。
+# 保护分支上禁止提交；暂存区卫生；快速 verify（.harness/engine/guards/git_guard.py）。
 . "$(dirname -- "$0")/_runtime"
-exec "$HARNESS_PY" "$HARNESS_GUARD" pre-commit
+exec "$HARNESS_PY" "$HARNESS_CLI" guard-git pre-commit

.githooks/pre-push

@@ -1,4 +1,4 @@
 #!/bin/sh
-# 禁推保护分支与 tag、禁强制推送；推送前完整 verify（harness/git_guard.py）。
+# 禁推保护分支与 tag、禁强制推送；推送前完整 verify（.harness/engine/guards/git_guard.py）。
 . "$(dirname -- "$0")/_runtime"
-exec "$HARNESS_PY" "$HARNESS_GUARD" pre-push "$@"
+exec "$HARNESS_PY" "$HARNESS_CLI" guard-git pre-push "$@"

.githooks/reference-transaction

@@ -1,5 +1,5 @@
 #!/bin/sh
-# 禁止本地改写/删除保护分支、移动/删除 tag（harness/git_guard.py）。
+# 禁止本地改写/删除保护分支、移动/删除 tag（.harness/engine/guards/git_guard.py）。
 # 每次引用更新都会调用本钩子：只在 prepared 阶段、且涉及分支或 tag 时才启动 Python。
 [ "$1" = prepared ] || exit 0
 input=$(cat)
@@ -10,4 +10,4 @@ esac
 # 解释器缺失时放行并告警：引用守卫失效不能让整个仓库无法提交。
 HARNESS_FAIL_OPEN=1
 . "$(dirname -- "$0")/_runtime"
-printf '%s\n' "$input" | exec "$HARNESS_PY" "$HARNESS_GUARD" reference-transaction prepared
+printf '%s\n' "$input" | exec "$HARNESS_PY" "$HARNESS_CLI" guard-git reference-transaction prepared

.github/pull_request_template.md

@@ -24,4 +24,4 @@
 ## 需要人工验收的部分
 
 <!-- 只列机器无法判定的项（真机 UI、系统授权等），写明步骤与预期；没有写「无」。
-     `python3 harness/acceptance.py --manual` 列出现役人工验收清单，从中挑出本 PR 涉及的编号。 -->
+     `bin/harness acceptance --manual` 列出现役人工验收清单，从中挑出本 PR 涉及的编号。 -->

.github/workflows/auto-merge.yml

@@ -1,6 +1,6 @@
 name: auto-merge
 
-# L4 自治：build 与 harness 全绿后，由 harness/policy.py 判定能否自动合并（目标态设计 9.3）：
+# L4 自治：build 与 harness 全绿后，由 .harness/engine/cli.py policy 判定能否自动合并（目标态设计 9.3）：
 # 风险 R0/R1、类别在 autonomy.toml 中为 L4、未超误差预算、不超规模阈值、任务 PR 的运行记录与 CI 轮次合格，
 # 每条理由写进 job summary。
 # 停机：Actions → auto-merge → Disable workflow；恢复：Enable workflow（docs/specs/delivery-harness.md §8）。
@@ -60,7 +60,7 @@ jobs:
           HEAD_SHA: ${{ github.event.workflow_run.head_sha }}
           PR: ${{ github.event.workflow_run.pull_requests[0].number }}
           HEAD_BRANCH: ${{ github.event.workflow_run.head_branch }}
-        run: python harness/policy.py --base origin/main --head "$HEAD_SHA" ${PR:+--pr "$PR"} --branch "$HEAD_BRANCH" --github
+        run: python .harness/engine/cli.py policy --base origin/main --head "$HEAD_SHA" ${PR:+--pr "$PR"} --branch "$HEAD_BRANCH" --github
 
   # 不自动合并的：build 与 harness 通过后打类别标签并请仓库所有者评审，PR 出现在其待评审列表里。
   # 用 GITHUB_TOKEN 只「请求评审」，不批准（仓库已关闭 Actions 批准 PR）。
@@ -145,5 +145,5 @@ jobs:
             gh label create audit --repo "$GITHUB_REPOSITORY" --color fbca04 --force >/dev/null
             gh issue create --repo "$GITHUB_REPOSITORY" --label audit --label "class:$CLASS" \
               --title "抽审：PR #$PR（$CLASS）" \
-              --body "PR #$PR 按 harness/autonomy.toml 的抽样规则被抽中（$RISK $CLASS，$HEAD_SHA）。评审方对照任务书、diff 与证据包审计；发现问题时另开带 escape 与 class:$CLASS 标签的议题，写明引入的 PR #$PR。"
+              --body "PR #$PR 按 .harness/config/autonomy.toml 的抽样规则被抽中（$RISK $CLASS，$HEAD_SHA）。评审方对照任务书、diff 与证据包审计；发现问题时另开带 escape 与 class:$CLASS 标签的议题，写明引入的 PR #$PR。"
           fi

.github/workflows/build.yml

@@ -42,45 +42,45 @@ jobs:
           fi
 
       - name: Verify (quick)
-        run: python harness/verify.py --quick
+        run: python .harness/engine/cli.py verify --quick
 
       - name: Hygiene of changes
-        run: python harness/hygiene.py --range "${{ steps.base.outputs.sha }}"
+        run: python .harness/engine/cli.py hygiene --range "${{ steps.base.outputs.sha }}"
 
       # 已有判定器以 base 为准：追加或新增的测试代码不能让已有测试失效（评审 PR7-R2）。
       - name: Existing tests (base version) on head
-        run: python harness/base_tests.py --base "${{ steps.base.outputs.sha }}"
+        run: python .harness/engine/cli.py base-tests --base "${{ steps.base.outputs.sha }}"
 
       # 前面的检查失败时，风险等级与证据仍要生成，评审才看得到全貌。
       - name: Risk level
         id: risk
         if: ${{ !cancelled() }}
-        run: python harness/risk.py --base "${{ steps.base.outputs.sha }}" --github
+        run: python .harness/engine/cli.py risk --base "${{ steps.base.outputs.sha }}" --github
 
       # R1 加强判定中要执行代码的一项（2026-09-28 决定 2）：声明 R1 时，改到的变异目标得分不得低于基线。
       # 下降即失败；要按 R2 走的重构去掉 Risk: R1 即可（加一个不带它的提交）。
       - name: Mutation score of changed targets (R1 claims)
         if: ${{ !cancelled() && steps.risk.outputs.claimed_r1 == 'true' }}
-        run: python harness/mutate.py --check --changed-since "${{ steps.base.outputs.sha }}"
+        run: python .harness/engine/cli.py mutate --check --changed-since "${{ steps.base.outputs.sha }}"
 
       # 实现任务书的 PR：归属 trailer 与运行记录（只报告；是否自动合并由 auto-merge 的 policy.py 判定）。
       - name: Run record (task PRs)
         if: ${{ !cancelled() }}
-        run: python harness/run_check.py --base "${{ steps.base.outputs.sha }}" --github
+        run: python .harness/engine/cli.py run-check --base "${{ steps.base.outputs.sha }}" --github
 
       - name: Fix evidence
         if: ${{ !cancelled() }}
-        run: python harness/evidence.py --base "${{ steps.base.outputs.sha }}" --markdown "$GITHUB_STEP_SUMMARY"
+        run: python .harness/engine/cli.py evidence --base "${{ steps.base.outputs.sha }}" --markdown "$GITHUB_STEP_SUMMARY"
 
       - name: Delivery metrics
         if: ${{ !cancelled() }}
         env:
           GITHUB_TOKEN: ${{ github.token }}
-        run: python harness/metrics.py --base "${{ steps.base.outputs.sha }}" --github --markdown "$GITHUB_STEP_SUMMARY"
+        run: python .harness/engine/cli.py metrics --base "${{ steps.base.outputs.sha }}" --github --markdown "$GITHUB_STEP_SUMMARY"
 
       - name: Release check (tag = build version, on main)
         if: startsWith(github.ref, 'refs/tags/v')
-        run: python harness/release_check.py --tag "$GITHUB_REF_NAME" --commit "$GITHUB_SHA"
+        run: python .harness/engine/cli.py release-check --tag "$GITHUB_REF_NAME" --commit "$GITHUB_SHA"
 
   build:
     # macos-26 自带 Xcode 26：build 脚本用 actool 把 native/AppIcon.icon 编成 Liquid Glass 分层图标。
@@ -105,7 +105,7 @@ jobs:
 
       - name: Verify (lint, hygiene, Python and Swift tests, incident replay)
         # 与本机、云端同一入口（docs/plans/verifiable-delivery.md P1）；--strict 不允许 Swift 检查被跳过。
-        run: scratch/iterm-probe-venv/bin/python harness/verify.py --strict --full
+        run: scratch/iterm-probe-venv/bin/python .harness/engine/cli.py verify --strict --full
 
       # Swift 测试的「修复前失败、修复后通过」只能在 macOS 上跑（harness job 在 Linux 只做 Python，方案 §13 E4）。
       - name: Fix evidence with Swift (pull requests)
@@ -114,7 +114,7 @@ jobs:
           BASE: ${{ github.event.pull_request.base.sha }}
         run: |
           git fetch --no-tags --unshallow origin || git fetch --no-tags origin
-          scratch/iterm-probe-venv/bin/python harness/evidence.py --base "$BASE" --swift --markdown "$GITHUB_STEP_SUMMARY"
+          scratch/iterm-probe-venv/bin/python .harness/engine/cli.py evidence --base "$BASE" --swift --markdown "$GITHUB_STEP_SUMMARY"
 
       - name: Build Agent Notification.app
         run: python3 scripts/build_inbox_app.py --standalone

.github/workflows/quality.yml

@@ -1,7 +1,7 @@
 name: quality
 
 # 每周的测试强度与熵治理检查（docs/specs/delivery-harness.md）：较慢，不放在每次提交的门禁里。
-# 周报（harness/weekly.py）在检查之后汇总交付指标，写进固定的「每周质量报告」议题；检查失败也照样出周报。
+# 周报（.harness/engine/cli.py weekly）在检查之后汇总交付指标，写进固定的「每周质量报告」议题；检查失败也照样出周报。
 on:
   schedule:
     - cron: '17 2 * * 1'  # 每周一 02:17 UTC
@@ -34,19 +34,19 @@ jobs:
         run: python -m pip install -r requirements-dev.txt
 
       - name: Mutation score not below baseline
-        run: python harness/mutate.py --check
+        run: python .harness/engine/cli.py mutate --check
 
       - name: Incident replay
         if: ${{ !cancelled() }}
-        run: python harness/replay.py
+        run: python .harness/engine/cli.py replay
 
       - name: Complexity ratchet
         if: ${{ !cancelled() }}
-        run: python harness/quality.py
+        run: python .harness/engine/cli.py quality
 
       - name: Weekly report
         if: ${{ !cancelled() }}
         env:
           GH_TOKEN: ${{ github.token }}
           GH_REPO: ${{ github.repository }}
-        run: python harness/weekly.py --publish | tee -a "$GITHUB_STEP_SUMMARY"
+        run: python .harness/engine/cli.py weekly --publish | tee -a "$GITHUB_STEP_SUMMARY"

.harness/config/checks.toml

@@ -0,0 +1,113 @@
+# delivery-harness 项目配置：检查清单、源码与语言、运行时、目录约定、身份。
+# 改动本文件属于 R3（护栏本身），由用户批准。
+schema = 1
+
+[identity]
+# 执行方与 Agent 推送、开 PR、评论使用的 GitHub 账号（bin/as-agent、派发的提交身份）。
+agent_login = "Snowson"
+agent_email = "15168246+Snowson@users.noreply.github.com"
+
+[runtime]
+# 开发包运行时 venv：bin/* 与 git 钩子优先用它，没有时退回 PATH 上的 python3。
+python = "scratch/iterm-probe-venv/bin/python"
+# 派发槽位清理未跟踪文件时保留、并以符号链接指向主目录的路径。
+preserve = ["scratch/iterm-probe-venv"]
+
+[sources]
+# 产品代码（R1 判定：签名不变、无新依赖、无迁移）。
+code = ["scripts/**", "native/**"]
+# UI 代码：policy 对 UI 改动不自动合并。
+ui = ["native/**"]
+# 质量棘轮：Python 用 ruff C901 与体量，Swift 用轻量解析。
+python_dirs = ["scripts"]
+swift_dirs = ["native"]
+# 数据迁移语句的检查只看产品代码。
+migrations = "scripts/migrations.py"
+
+[release]
+# 发版前核对：tag 名 = "v" + 版本号；两处版本号必须同步。
+version_file = "scripts/build_inbox_app.py"
+short_version_var = "BUNDLE_SHORT_VERSION"
+build_version_var = "BUNDLE_VERSION"
+
+[[mutation.targets]]
+name = "价格转换与防护"
+file = "scripts/pricing_fetch.py"
+functions = ["transform", "_price_jump", "_drop_conflicting_aliases"]
+tests = ["test_pricing_fetch", "test_properties.PricingTransformProperties"]
+
+[[mutation.targets]]
+name = "热力金额分级"
+file = "scripts/daily_report.py"
+functions = ["cost_thresholds", "cost_level"]
+tests = ["test_properties.CostThresholdProperties", "test_daily_report"]
+
+[[mutation.targets]]
+name = "过去日报告合并"
+file = "scripts/daily_report.py"
+functions = ["_merge_day_tasks", "_task_total"]
+tests = ["test_properties.MergeNoDowngradeProperties", "test_daily_report"]
+
+[[mutation.targets]]
+name = "模型规范名"
+file = "scripts/model_names.py"
+functions = ["canonical"]
+tests = ["test_model_names"]
+
+[[mutation.targets]]
+name = "金额计算"
+file = "scripts/usage_cost.py"
+functions = ["price_for_day", "model_cost", "cost_for_models", "merge_cost", "convert_display"]
+tests = ["test_usage_cost"]
+
+[[mutation.targets]]
+name = "用量汇总"
+file = "scripts/usage_report.py"
+functions = ["period_bounds", "previous_anchor", "_merge_models", "_dimension_rows"]
+tests = ["test_usage_report", "test_golden"]
+
+[verify]
+requirements = "requirements-dev.txt"
+pinned_tools = ["ruff"]
+order = [
+    "tools", "integrity", "lint", "hygiene", "quality", "docs", "acceptance", "taskbook",
+    "python-tests", "swift-policy", "swift-runner", "zcode-selftest", "replay",
+]
+
+[[verify.checks]]
+name = "lint"
+tiers = ["quick", "default", "full"]
+command = ["{python}", "-m", "ruff", "check", "scripts", "tests"]
+
+[[verify.checks]]
+name = "python-tests"
+command = ["{python}", "-W", "error::ResourceWarning", "-m", "unittest", "discover", "-s", "tests"]
+
+[[verify.checks]]
+name = "swift-policy"
+shell = "mkdir -p build && xcrun swiftc -parse-as-library native/InboxPolicy.swift native/Shared/WidgetSnapshot.swift tests/InboxPolicyTests.swift -o build/policy-tests && ./build/policy-tests"
+requires = "macos"
+why_skipped = "非 macOS，Swift 检查由 macOS CI 负责"
+
+[[verify.checks]]
+name = "swift-runner"
+shell = "mkdir -p build && xcrun swiftc -parse-as-library native/ProcessRunner.swift tests/ProcessRunnerTests.swift -o build/runner-tests && ./build/runner-tests"
+requires = "macos"
+why_skipped = "非 macOS，Swift 检查由 macOS CI 负责"
+
+[[verify.checks]]
+# AGENTS.md 要求的无 UI 自检（身份比对、搜索框结构、点击状态机、run loop）。
+name = "zcode-selftest"
+shell = "mkdir -p build && xcrun swiftc native/ZcodeFocus.swift -o build/zcode-focus && build/zcode-focus --self-test"
+requires = "macos"
+why_skipped = "非 macOS，Swift 检查由 macOS CI 负责"
+
+[replay]
+# 回放用例中 tests 写 "swift-policy" 时运行的命令（macOS）。
+[replay.suites.swift-policy]
+shell = "mkdir -p build && xcrun swiftc -parse-as-library native/InboxPolicy.swift native/Shared/WidgetSnapshot.swift tests/InboxPolicyTests.swift -o build/policy-tests && ./build/policy-tests"
+requires = "macos"
+
+[reports]
+# 自动合并用的批准 App 的机器人账号（周报把它合并或批准的 PR 算作自动合并）。
+bots = ["snowsonz-auto-merge"]