hongzuoj-pixel · Oct 19, 2026 · Oct 19, 2026 · Oct 19, 2026
diff --git a/.github/workflows/verification.yml b/.github/workflows/verification.yml
@@ -216,14 +216,27 @@ jobs:
           mimo --config examples/cli_agent/mimo-harness.json report --run examples/cli_agent/.mimo/runs/recovered-repair
           mimo export --run examples/cli_agent/.mimo/runs/recovered-repair --output cli-repair-evidence.zip
           mimo gate --bundle cli-repair-evidence.zip
+      - name: Prepare portable user-kit dependencies and rehearse clean installation
+        run: |
+          python -m pip download --only-binary=:all: --platform any --implementation py --abi none --dest .mimo/kit-wheels pip==25.0.1 pytest==8.4.2 iniconfig==2.1.0 packaging==25.0 pluggy==1.6.0 pygments==2.19.2 exceptiongroup==1.3.0 tomli==2.2.1 colorama==0.4.6 typing_extensions==4.15.0
+          python -c "import pathlib,shutil; shutil.copy2(next(pathlib.Path('.mimo/dist').glob('*.whl')), '.mimo/kit-wheels')"
+          python tests/ci_user_kit.py --wheelhouse .mimo/kit-wheels --output .mimo/user-kit-rehearsal
       - uses: actions/upload-artifact@v4
         if: always()
         with:
           name: cli-pytest-${{ matrix.os }}
           path: |
             cli-pytest-results.xml
             .mimo/dist/*
+            .mimo/user-kit-rehearsal/kit/
+            .mimo/user-kit-rehearsal/rehearsal.json
           include-hidden-files: true
+      - uses: actions/upload-artifact@v4
+        with:
+          name: independent-user-kit-${{ matrix.os }}
+          path: .mimo/user-kit-rehearsal/kit/
+          include-hidden-files: true
+          if-no-files-found: error
 
   unit-and-smoke:
     runs-on: ubuntu-latest

diff --git a/README.md b/README.md
@@ -258,6 +258,8 @@ mimo readiness
 
 `readiness` returns `3` while external acceptance is pending. Use `trial-init`/`trial-run` for an explicitly budgeted live model trial and `walkthrough-init` for each independent new user. Blank templates and deterministic controls do not count as those observations. See the [acceptance guide](docs/PRODUCT_ACCEPTANCE_CN.md) and [support/upgrade boundaries](docs/SUPPORT_AND_UPGRADES.md).
 
+v0.12 adds [`walkthrough-pack`](docs/INDEPENDENT_USER_KIT_CN.md): distribute one exact wheel, offline Python dependencies and three blank user-study slots. A cross-platform bootstrap creates a private environment, verifies the installed engine and preserves feedback when resuming interrupted setup. Actual user observations remain separate.
+
 ### Continue an interrupted run
 
 ```bash

diff --git a/README_CN.md b/README_CN.md
@@ -215,6 +215,8 @@ v0.11 已增加原生 Node.js 与通用 JUnit 的逐用例验证，可通过 `ms
 
 真实模型试跑、新用户反馈和最终验收分别使用 `trial-init`/`trial-run`、`walkthrough-init` 和 `readiness`。执行 `mimo readiness` 可查看缺项；返回 `3` 表示尚未齐备，不会把测试模板当作真实试用。详见[中文产品验收指南](docs/PRODUCT_ACCEPTANCE_CN.md)和[支持范围与升级说明](docs/SUPPORT_AND_UPGRADES.md)。
 
+v0.12 新增 [`walkthrough-pack` 独立用户试用包](docs/INDEPENDENT_USER_KIT_CN.md)：三位使用者共用同一份 wheel 和离线 Python 依赖，用一条命令创建独立环境；安装中断后可以恢复，已填写的反馈会保留。自动安装不算真人验收。
+
 ### 6.3 初始化并检查项目
 
 先使用明确的离线模式。下面命令会创建 `mimo-harness.json` 和 `.mimo/models.json`，不会调用模型 API。

diff --git a/docs/ACCEPTANCE_MATRIX.md b/docs/ACCEPTANCE_MATRIX.md
@@ -59,7 +59,7 @@ The two actual-Codex tests require `MIMO_CODEX_BINARY` and the exact pinned exec
 - Signed/remote attestation, richer evidence migration and organization-level enforced workflow policies are outside the current unsigned bundle format. The opt-in PR template still requires maintainer-configured required checks/review protection.
 - Generic CLI Agent repair, the pinned Codex event adapter and two real Python defect controls have automated gates. Live paid-provider/model repair validation still requires credentials and an explicit spending allowance. Opaque external Agent tokens/costs remain unavailable; invocation/time caps are not billing caps.
 - Broader language/framework and defect diversity beyond the documented native Node/JUnit subset; Keil/firmware/hardware behavior remain separately scoped.
-- Maintainer-selected license and at least three independent new-user walkthroughs. Wheel/sdist checks, upgrade/support notes, trial/feedback generation and source-bound CI collection are implemented; generated templates are not completed external evidence.
+- Maintainer-selected license and at least three independent new-user walkthroughs. Wheel/sdist checks, upgrade/support notes, exact-wheel user kits, offline install/setup recovery, trial/feedback generation and source-bound CI collection are implemented; generated templates and automated installer rehearsals are not completed external evidence.
 - One final clean checkout/install run across the supported matrix, all failure-injection scenarios, evidence audit and scope review after the last functional change.
 
 Passing an intermediate batch is not the final completeness sign-off. Unavailable environments, skipped tests and external/manual gates must remain visible in the release record.

diff --git a/docs/INDEPENDENT_USER_KIT_CN.md b/docs/INDEPENDENT_USER_KIT_CN.md
@@ -0,0 +1,40 @@
+# 把同一份可验证安装包交给三位新用户
+
+v0.12 beta 提供 `walkthrough-pack`，把当前引擎对应的 wheel、固定依赖、跨平台安装脚本、中文指南和三份空白反馈放在同一个目录。三位使用者安装完全相同的引擎文件，避免各自重建 wheel 带来身份差异。生成材料或自动演练不计为真人试用。
+
+## 维护者准备
+
+在准备交付的干净 checkout 中构建并安装 wheel：
+
+```bash
+python -m pip install build==1.3.0
+python -m build --outdir .mimo/dist
+python -m pip install --force-reinstall --no-deps .mimo/dist/mimo_codeharness-0.12.0-py3-none-any.whl
+python -m pip download --only-binary=:all: --platform any --implementation py --abi none --dest .mimo/kit-wheels pip==25.0.1 pytest==8.4.2 iniconfig==2.1.0 packaging==25.0 pluggy==1.6.0 pygments==2.19.2 exceptiongroup==1.3.0 tomli==2.2.1 colorama==0.4.6 typing_extensions==4.15.0
+python -c "import shutil; shutil.copy2('.mimo/dist/mimo_codeharness-0.12.0-py3-none-any.whl', '.mimo/kit-wheels')"
+mimo walkthrough-pack --wheelhouse .mimo/kit-wheels --output .mimo/user-kit
+```
+
+准备依赖需要联网；输出目录必须是新目录。wheel 必须与当前 CLI 的版本和引擎字节一致；缺依赖、重复版本、平台专用 wheel 和不同引擎都会被拒绝。安装时使用这些确切 wheel，不从在线索引解析新依赖。
+
+CI 的 Linux/Windows `cli-install` 任务从真实构建产物生成试用包，并演练新环境安装、拒绝覆盖、半成品虚拟环境恢复、保留反馈、真实 boltons 对照验收、手动恢复和迁移后的 ZIP 门禁。选择一份可信 CI 的 `independent-user-kit-ubuntu-latest` 或 `independent-user-kit-windows-latest` artifact 下载；解压后即可看到 bootstrap.py 和 GUIDE_CN.md。原样分发同一份包给三位使用者，不要混合不同任务中重建的 wheel。技术演练记录 rehearsal.json 另保留在 `cli-pytest-*` artifact 中。
+
+## 新用户操作
+
+使用者只需预装 Python 3.10+ 和 Git，在试用包目录运行：
+
+```text
+python bootstrap.py --participant user-01 --output my-study
+```
+
+另外两人选择 user-02、user-03。这些名称只是材料分配槽位，真实匿名 tester_id 由使用者自己填写。
+
+安装脚本先校验传输文件，在新目录创建 `.venv`，仅离线安装包内依赖并核对实际安装的引擎；不修改系统 Python，不需要激活环境。随后进入 my-study，按照 GUIDE_CN.md 使用 `python mimo.py ...` 完成流程。Git 下载固定上游源码需要访问 GitHub，离线安装不等于整个真实案例都不需要网络。
+
+安装失败或中断后，使用相同参数加 `--resume`。每次尝试保留独立安装日志，技术安装状态写入 setup.json。已有 feedback.json 和报告不会被自动填写或覆盖；无法恢复时保留原目录并使用新输出目录。
+
+试用者提交自己填写的 feedback.json、自己运行生成的 evidence.zip 及必要日志。反馈仍要经过 [readiness](PRODUCT_ACCEPTANCE_CN.md) 检查和人工独立性复核。包中哈希和 CI JSON 均未签名，用于检查版本/传输一致性；请从可信维护者和 CI 取得材料。
+
+汇总验收的维护者也应安装这份包中的同一 wheel，再运行 readiness。仅版本号相同、分别重建的代码文件未必具有相同的引擎身份。
+
+本流程不调用模型、不设置模型额度，也不替维护者选项目许可证。[真实模型和最终发布验收](PRODUCT_ACCEPTANCE_CN.md) 仍单独进行。
diff --git a/docs/PRODUCT_ACCEPTANCE_CN.md b/docs/PRODUCT_ACCEPTANCE_CN.md
@@ -33,6 +33,8 @@ mimo trial-run --plan .mimo/live-trial-plan.json --output .mimo/live-trial-001
 
 ## 2. 至少三位独立新用户试用
 
+推荐先按[独立用户试用包指南](INDEPENDENT_USER_KIT_CN.md)准备同一份 wheel 和离线依赖，让三位使用者获得一致的引擎。下面的 `walkthrough-init` 仍可用于已经安装好相同版本的环境。
+
 为每位未参与实现的使用者生成一份独立材料：
 
 ```bash

diff --git a/docs/PRODUCT_PLAN.md b/docs/PRODUCT_PLAN.md
@@ -16,7 +16,7 @@ Priorities are ordered by user dependency, with a concrete exit condition:
 | P2b · adapter implemented; live trial pending | Ready-to-use vendor adapter, then live model validation | Pinned Codex CLI adapter and config generator; actual-binary/local-server CI. A real model still requires available credentials, an explicit spending allowance and a separately recorded outcome |
 | P3a · implemented | Two independent real Python defect cases | boltons and more-itertools pinned source/fix commits; reproduced target assertions, reference acceptance, regression rejection, fresh recovery and portable evidence via `case-init`/`case-check` |
 | P3b · implemented | Native Node and generic JUnit case checks | Actual Node 22/24, typed JUnit contract, ms/boltons reference and regression controls, fixed acceptance, fresh recovery and exact portable proof; no zero-exit-only PASS |
-| P4 · workflow prepared; real users pending | New-user workflow and actionable diagnosis | Chinese guide and blank feedback generation, duplicate/stale/unfinished evidence checks. At least three actual independent new users still need to complete the workflow and resolve observed issues |
+| P4 · workflow prepared; real users pending | New-user workflow and actionable diagnosis | Exact-engine user kit, offline dependency installation, interrupted setup recovery, Chinese guides and three blank feedback slots; duplicate/stale/unfinished evidence checks. At least three actual independent new users still need to complete the workflow and resolve observed issues |
 | P5 · automation/docs prepared; final sign-off pending | Release and final completeness sign-off | Wheel/sdist, support/upgrade notes, exact-source CI record and readiness gate. Maintainer license choice, live trial, independent users and final manual review remain external gates |
 
 P2–P4 may be scoped into smaller batches, but a capability is not advertised as certified before its exit condition is met. Existing `mimo verify` already accepts patches produced by external tools without needing a tool-specific Agent integration.

diff --git a/docs/SUPPORT_AND_UPGRADES.md b/docs/SUPPORT_AND_UPGRADES.md
@@ -1,6 +1,6 @@
 # Supported scope and upgrades
 
-v0.11 remains a beta. Automated release evidence and a stable release sign-off are separate. The package has no mandatory third-party Python runtime dependencies; configured projects/runners need their own explicitly prepared dependencies.
+v0.12 remains a beta. Automated release evidence and a stable release sign-off are separate. The package has no mandatory third-party Python runtime dependencies; configured projects/runners need their own explicitly prepared dependencies.
 
 | Component | Automated coverage | Boundary |
 |---|---|---|
@@ -12,6 +12,7 @@ v0.11 remains a beta. Automated release evidence and a stable release sign-off a
 | Linux Docker | Actual non-root isolated baseline/candidate/Agent tests; Node/JUnit included | Prebuilt image, exact resolved image ID, explicit network/resources; no local fallback |
 | Codex | CLI 0.163.0 native-binary protocol on Linux/Windows | Explicit model/key; live model/tool quality and billing require observed trials |
 | Portable evidence | Relocation, strict gate, HTML/JSON/JUnit/log linkage; exact Node/JUnit identity | Unsigned integrity evidence, not hostile-party authentication |
+| Independent user kit | One exact wheel, portable offline Python dependencies; actual Python 3.12 installation/recovery on Linux/Windows | Python/Git must already be installed; fetching pinned upstream cases needs GitHub; automated rehearsal is not a human trial |
 | Embedded/other languages | Existing scans and explicit commands | No claim of full Keil firmware build, hardware validity or certified Go/Rust/Jest/Vitest acceptance |
 
 ## Install and upgrade
@@ -22,7 +23,7 @@ Keep the previous checkout/tag and all run directories before upgrading. Do not
 
 Recovery requires the exact source/test/configuration, task, engine, executable/environment and Docker identity that produced the original run. A version upgrade, changed reporter or changed Node binary can invalidate recovery. Restore the matching old environment to resume an old run, or start a new run; do not edit fingerprints/accounting to force reuse. Previously completed portable bundles remain available for inspection, subject to their evidence format and verifier compatibility.
 
-v0.11 does not migrate cached tests or silently reuse old PASS. Resume recreates baseline/candidate checks; completed proposals are reused only when inputs match. Native Node executable bytes are included in local recovery identity; Docker runtime bytes are bound through image identity.
+The current version does not migrate cached tests or silently reuse old PASS. Resume recreates baseline/candidate checks; completed proposals are reused only when inputs match. Native Node executable bytes are included in local recovery identity; Docker runtime bytes are bound through image identity.
 
 ## Final release record
 

diff --git a/pyproject.toml b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mimo-codeharness"
-version = "0.11.0"
+version = "0.12.0"
 description = "Evidence-backed repository and code-agent patch verification harness"
 readme = "README.md"
 requires-python = ">=3.10"

diff --git a/src/mimo_harness/__init__.py b/src/mimo_harness/__init__.py
@@ -1,4 +1,4 @@
 """Installable command-line interface for MiMo-CodeHarness."""
 
-__version__ = "0.11.0"
+__version__ = "0.12.0"
 
diff --git a/src/mimo_harness/cli.py b/src/mimo_harness/cli.py
@@ -1017,6 +1017,11 @@ def build_parser() -> argparse.ArgumentParser:
     walkthrough = sub.add_parser("walkthrough-init", help="Create an independent-user guide and blank feedback record.")
     walkthrough.add_argument("--output", required=True)
     walkthrough.set_defaults(handler=command_walkthrough_init)
+    from mimo_harness.user_kit import command_walkthrough_pack
+    pack = sub.add_parser("walkthrough-pack", help="Package one exact wheel and offline dependencies for three independent user walkthroughs.")
+    pack.add_argument("--wheelhouse", required=True, help="Directory containing the matching harness wheel and all portable pytest dependencies")
+    pack.add_argument("--output", required=True, help="New kit directory; existing files are preserved")
+    pack.set_defaults(handler=command_walkthrough_pack)
     readiness = sub.add_parser("readiness", help="Check source-bound product evidence; missing external gates remain pending.")
     readiness.add_argument("--repository", default=".")
     readiness.add_argument("--trial", help="Completed trial.json")

diff --git a/src/mimo_harness/manifests.py b/src/mimo_harness/manifests.py
@@ -182,7 +182,7 @@ def environment_identity(execution: dict, checks, salt: str, runtime_home: Path)
             "environment_sha256": digest([salt, env])}
 
 
-def engine_identity() -> str:
+def engine_inventory() -> dict:
     # A wheel lives in site-packages beside unrelated packages; a checkout may
     # contain legacy scripts that are not distributed. Hash the same executable
     # components in both layouts, using module names instead of install paths.
@@ -191,7 +191,11 @@ def engine_identity() -> str:
     for name in ("run_mimo_codeharness_v02", "scoring_v03", "verification_v04"):
         spec = importlib.util.find_spec(name)
         code[name + ".py"] = file_digest(Path(spec.origin)) if spec and spec.origin else None
-    return digest(code)
+    return code
+
+
+def engine_identity() -> str:
+    return digest(engine_inventory())
 
 
 class RecoverySource:

diff --git a/src/mimo_harness/product_acceptance.py b/src/mimo_harness/product_acceptance.py
@@ -133,7 +133,7 @@ def command_walkthrough_init(args):
 3. 执行 `mimo --config .mimo/user-case/mimo-harness.json doctor`。
 4. 执行 `mimo case-check --directory .mimo/user-case --name first`。阅读参考 PASS、故意回归 FAIL 和恢复重跑的证据；整个对照验收应成功。
 5. 再用相同名称运行一次，确认拒绝覆盖；记录提示是否足以解决问题。换一个新名称可重新验收。
-6. 将 `results/first/reference.zip` 复制到其他目录，执行 `mimo gate --bundle ZIP路径 --json`；解压后打开 report.html。
+6. 将 `.mimo/user-case/results/first/reference.zip` 复制到其他目录，执行 `mimo gate --bundle ZIP路径 --json`；解压后打开 report.html。
 7. 填写匿名 tester_id、观察时间、环境、各步骤是否完成、遇到的问题。只在实际独立试用后设置 independence_confirmed=true。issues 中每个问题记录 description 和 resolved。
 8. evidence_bundle 填写你自己运行产生的参考 ZIP 路径（相对 feedback.json 所在目录，或绝对路径）。不同试用者需要各自的运行证据。
-Original file line number
+Diff line change
@@ Expand Up / @@ -258,6 +258,8 @@ mimo readiness @@
     `readiness` returns `3` while external acceptance is pending. Use `trial-init`/`trial-run` for an explicitly budgeted live model trial and `walkthrough-init` for each independent new user. Blank templates and deterministic controls do not count as those observations. See the [acceptance guide](docs/PRODUCT_ACCEPTANCE_CN.md) and [support/upgrade boundaries](docs/SUPPORT_AND_UPGRADES.md).
+    v0.12 adds [`walkthrough-pack`](docs/INDEPENDENT_USER_KIT_CN.md): distribute one exact wheel, offline Python dependencies and three blank user-study slots. A cross-platform bootstrap creates a private environment, verifies the installed engine and preserves feedback when resuming interrupted setup. Actual user observations remain separate.
     ### Continue an interrupted run
     ```bash
@@ Expand Down @@