owlrunkit 0.17.2 → 0.18.1

package/CHANGELOG.md

@@ -1,5 +1,69 @@
 # Changelog
 
+## 0.18.1
+
+- Adds read-only `mode recommend` and `mode status` commands with five
+  user-facing presets over independent continuity and assurance dimensions.
+- Routes long-running, interruption recovery, and ordinary reversible project
+  mutation to Project Driver continuity without automatically forcing Formal.
+  Multi-writer, permission/security, funds, irreversible, rollback-uncertain,
+  schema/data migration, and explicit acceptance facts still force Formal.
+- Fixes Quick metrics to report receipt-time pass, fail, and source mutation
+  independently of later workspace drift.
+- Returns the standalone `quick-attest` command from Quick results and adds
+  action-specific `quick-attest --help`.
+- Diagnoses managed Skill/config Core version drift independently from installed
+  file integrity.
+- Aligns `doctor` with full profile validation, stops profile detection from
+  inventing absent `src/**` and `tests/**` surfaces, and adds atomic profile
+  reconciliation for an existing project.
+- Adds action-specific `adopt --help` and keeps top-level inspection focused on
+  the Project Driver's dominant project gap when one exists.
+- Preserves `authorizationGranted=false` and the separation between workflow,
+  acceptance, Git, release, deployment, production, and business authority.
+
+## 0.18.0
+
+- Introduces the Project Driver product path for continuity across Agents and
+  sessions, with typed direct `init`, `assign`, `checkpoint`, `handoff`,
+  `decision`, `verification`, `integrate`, `status`, `takeover`, and
+  bounded `successor` actions.
+- Projects a deterministic dominant gap, next action, next actor, and headline
+  from durable project artifacts without accepting guessed percentages or
+  inventing completion.
+- Keeps legacy `--request` event input for compatibility while removing the
+  need for normal users to author request, event, receipt, or hash JSON.
+- Clarifies that ordinary multi-Agent coordination does not automatically mean
+  Formal; the assurance boundary follows the underlying risk and authority.
+- Adds a deferred-verification ledger that records stable unrun check IDs,
+  reason, owner, and due integration gate without executing tests. Due open
+  checks block only their declared gate; closure requires verification evidence
+  or an already resolved project decision that says the check is no longer
+  required.
+- Keeps coverage reuse in the existing verification-plan/receipt substrate, so
+  Project Driver status exposes the remaining gap instead of creating a second
+  test engine or repeatedly rerunning covered checks.
+- Defines the 0.18 Project Driver as a bounded surface for new projects and
+  controlled dogfood. Its product name does not claim to be a general
+  long-running, multi-phase, or enterprise control plane.
+- Adds an Owner-invoked Project Successor for one derived-completed active V1
+  project: archive the exact prior bytes under a content-addressed identity,
+  install one different empty V1 project at the same active path, and recover
+  the atomic transition from a durable journal after interruption.
+- Rejects incomplete or changed project truth, the same project identity,
+  symlinked inputs, archive collisions, and competing successor transactions.
+  Completed archives are sealed against ordinary writes; exact retry and the
+  next successor boundary validate every completed journal's archive bytes and
+  seal before active truth or a new journal can change. Status and takeover do
+  not perform that historical scan. Incomplete projects stay on their current
+  truth. The receipt grants no Git, release, publication, deployment, or
+  adoption authority. Handoff and takeover do not automatically promote an
+  authoritative writer.
+- Keeps Core successor, release compatibility, publication, Project Successor,
+  and project adoption as separate authority boundaries.
+- Preserves the published `0.17.2` Core and package as the trusted rollback
+  prior for the independent `0.18.0` candidate.
+
 ## 0.17.2
 
 - Treats valid profile launcher portability warnings as non-blocking maintenance,

package/README.md

@@ -1,14 +1,31 @@
 # OwlRunKit
 
-OwlRunKit is OwlCoda's project-owned execution and evidence engine, released
-with its own version lifecycle. It makes the correct workflow the ordinary
-path: choose the assurance level, run bounded work, preserve source and
-evidence, then close out once.
+OwlRunKit is OwlCoda's Project Driver for cross-Agent project continuity and
+progression. Its execution, evidence, verification, and recovery substrate
+makes that continuity inspectable and recoverable. RunKit 0.18 keeps work
+moving across Agents and sessions through a typed direct flow while preserving
+the existing assurance, source, and evidence boundaries.
+
+RunKit 0.18.1 adds a human-first usage-mode recommendation that selects
+continuity and assurance independently. It keeps ordinary development light
+while preserving mandatory Formal routing for material safety and authority
+facts.
+
+The 0.18 Project Driver is a bounded surface for new projects and controlled
+dogfood. Its product name does not claim to be a general long-running,
+multi-phase, or enterprise control plane. Version 0.18 adds one bounded
+Project Successor operation: an Owner may archive the exact prior bytes of a
+derived-completed active project and install a different, empty V1 project as the sole
+active project. It is not a general migration or multi-project writer system.
+Incomplete projects stay on their current truth. Handoff and takeover do not
+automatically promote an authoritative writer. Core successor, release
+compatibility, publication, Project Successor, and project adoption remain
+separate authority boundaries.
 
 ## Install
 
 ```bash
-npm install --save-exact owlrunkit@0.17.2
+npm install --save-exact owlrunkit@0.18.1
 ```
 
 The `owlrunkit` bootstrap uses the project's exact locally bound CLI for
@@ -24,24 +41,110 @@ the project Config must independently bind the same Core.
 owlrunkit --version
 owlrunkit init --workspace /absolute/path/to/project
 owlrunkit doctor --workspace /absolute/path/to/project
+owlrunkit mode recommend --workspace /absolute/path/to/project
+owlrunkit mode status --workspace /absolute/path/to/project
 ```
 
-## Coordinate a multi-Agent project
+`mode recommend` is read-only and defaults to human output; add `--json` for a
+machine contract. The five user modes are presets over `continuity`
+(`none | project_driver`) and `assurance` (`none | quick | formal`). They never
+grant Git, release, deployment, production, or business authority.
 
-RunKit 0.17 keeps the project alive independently of any one Agent or chat.
-Define durable milestones, workstreams, WorkItems, dependencies, integration
-gates, and optional measured units, then record append-only assignment and
-checkpoint events:
+## Project Driver: Coordinate a multi-Agent project
+
+RunKit 0.18 keeps the project alive independently of any one Agent or chat.
+Initialize a definition, then use typed direct commands for the normal
+assignment, checkpoint, handoff, decision, integration, status, and takeover
+flow:
 
 ```bash
+cp node_modules/owlrunkit/scripts/runkit-contract/templates/team-project-definition.json ./project-definition.json
 owlrunkit project init --workspace . \
-  --definition scripts/runkit-contract/templates/team-project-definition.json
-owlrunkit project assign --workspace . \
-  --request scripts/runkit-contract/templates/team-project-event.json
+  --definition ./project-definition.json
+owlrunkit project assign --workspace . --assignment-id assign-1 \
+  --at 2026-08-09T00:00:00.000Z --work-item work-item-1 --agent agent-runtime
+owlrunkit project checkpoint --workspace . --checkpoint-id checkpoint-1 \
+  --at 2026-08-09T00:05:00.000Z --assignment-id assign-1 --work-item work-item-1 \
+  --state active --summary 'Implement the bounded work item' \
+  --completed-units 0 \
+  --next 'Run the focused verification'
+owlrunkit project handoff --workspace . --handoff-id handoff-1 \
+  --at 2026-08-09T00:10:00.000Z --assignment-id assign-1 --work-item work-item-1 \
+  --from-agent agent-runtime --to-agent agent-successor \
+  --summary 'Source work is ready for continuation' --next 'Run verification' \
+  --evidence evidence/source.txt
+owlrunkit project assign --workspace . --assignment-id assign-2 \
+  --at 2026-08-09T00:11:00.000Z --work-item work-item-1 \
+  --agent agent-successor --supersedes assign-1
+owlrunkit project decision --workspace . --open --decision-id decision-1 \
+  --at 2026-08-09T00:12:00.000Z --title 'Choose the accepted source path' \
+  --question 'Which source path should continue?' --owner-agent agent-successor \
+  --blocking-work-item work-item-1 --option accepted --option rejected
+owlrunkit project decision --workspace . --resolve --decision-id decision-1 \
+  --at 2026-08-09T00:13:00.000Z --resolution 'Use the accepted source path' \
+  --rationale 'It preserves the project contract' --evidence evidence/decision.txt
+owlrunkit project checkpoint --workspace . --checkpoint-id checkpoint-2 \
+  --at 2026-08-09T00:14:00.000Z --assignment-id assign-2 \
+  --work-item work-item-1 --state completed --completed-units 1 \
+  --summary 'Accepted work is complete' --evidence evidence/completion.txt
+owlrunkit project verification --workspace . --defer \
+  --verification-id release-batch --at 2026-08-09T00:14:30.000Z \
+  --work-item work-item-1 --owner-agent agent-successor \
+  --check typecheck --check full-suite --check package-smoke \
+  --reason 'Run broad checks once after the related changes settle' \
+  --due-gate integration-ready
+# After those checks run once against the frozen candidate:
+owlrunkit project verification --workspace . --close \
+  --verification-id release-batch --at 2026-08-09T00:14:45.000Z \
+  --disposition verified --summary 'The batched release checks passed' \
+  --evidence evidence/release-batch.txt
+owlrunkit project integrate --workspace . --gate integration-ready \
+  --at 2026-08-09T00:15:00.000Z --summary 'Integration gate evidence is complete' \
+  --evidence evidence/integration.txt
 owlrunkit project status --workspace .
-owlrunkit project takeover --workspace . --agent agent-runtime
+owlrunkit project takeover --workspace . --agent agent-successor
+owlrunkit project successor --workspace . --transition-id project-next-001 \
+  --at 2026-08-09T00:16:00.000Z --definition ./next-project-definition.json \
+  --reason 'Archive the completed project and begin the next bounded project'
 ```
 
+Normal users supply the one definition JSON to `init`, then use typed flags and
+values. They do not need to author request, event, receipt, or hash JSON. The
+legacy `--request <event.json>` path remains available for existing automation
+and compatibility. Multiple Agents do not make project coordination Formal by
+themselves. This flow does not make project coordination Formal by itself; the
+assurance lane follows the underlying high-risk integration,
+release, production, migration, destructive, or explicit Formal-acceptance boundary.
+
+`project verification --defer` is a small deferred-verification ledger, not a
+test scheduler. It records stable check IDs, why they were intentionally not
+run yet, the responsible Agent, and the integration gate before which they
+must close. `status` and `takeover` keep that gap visible without repeatedly
+executing anything; the due gate stays blocked while the record is open.
+Closing as `verified` requires evidence. Closing a check that is no longer
+required uses `--disposition no_longer_required` and must reference an already
+resolved project decision. RunKit never turns either disposition into Git,
+release, deployment, or business authority.
+
+The existing coverage-adoption and verification-plan surfaces remain the
+source for reusable receipts and exact pending commands in the Formal lane.
+The Project Driver records the remaining coordination gap; it does not create
+a second test engine or ask an Agent to repeat already covered checks.
+
+`project successor` is available only after the current project derives
+`overall=completed`. It keeps the active path at
+`.owlcoda/runkit/project`, moves the exact prior bytes into a
+content-addressed `project-archives` entry, and installs a new definition with
+an empty event directory. A durable transition journal makes the same request
+restartable after interruption; a competing transition, symlink, archive
+collision, changed truth, same project identity, or incomplete project fails
+closed. Archived regular files and directories are sealed against ordinary
+writes with the receipt-bound `0444`/`0555` policy before completion. Archive
+integrity and that seal are validated on an exact retry and before any later
+successor; `project status` and `project takeover` remain active-truth
+projections and do not perform a broad historical archive scan. The receipt
+always keeps Git, release, publish, deploy, and adoption authority false.
+
 `project status` reports every Agent's active, waiting, completed, and failed
 work; unresolved dependencies and decisions; the ready queue; critical work;
 and integration gates. It does not accept a guessed completion percentage.

package/docs/architecture/OWLCODA_RUNKIT_017_AGENT_NATIVE_PRODUCT_DEFINITION.md

@@ -1,6 +1,6 @@
 # OwlRunKit 0.17 Agent-Native Product Definition
 
-Status: implemented candidate pending Formal acceptance and public release.
+Status: released in `owlrunkit@0.17.2` and integrated into the private mainline.
 
 ## Product purpose
 
@@ -150,5 +150,4 @@ The release is acceptable only when all of these are mechanically demonstrated:
 9. existing 0.16 execution, migration, fleet, deployment, and release contracts
    remain compatible;
 10. the standalone npm artifact is reproducible, scrubbed, identity-bound, and
-    independently verified after publication.
-
+   independently verified after publication.

package/docs/architecture/OWLCODA_RUNKIT_018_PROJECT_DRIVER_PRODUCT_DEFINITION.md

@@ -0,0 +1,165 @@
+# OwlCoda RunKit 0.18 Project Driver Product Definition
+
+Status: public product contract for the standalone `owlrunkit@0.18.0` candidate.
+This document defines the normal project-continuity path; it does not grant Git,
+release, npm, deployment, production, migration, destructive, or business
+authority.
+
+The 0.18 Project Driver is a bounded surface for new projects and controlled
+dogfood. Its product name does not claim to be a general long-running,
+multi-phase, or enterprise control plane. Version 0.18 includes one bounded
+Project Successor lifecycle operation for a derived-completed active V1
+project. It is not a general existing-project migration, multi-project
+registry, or automatic writer-promotion system. Incomplete projects stay on
+their current truth. Handoff and takeover do not automatically promote an
+authoritative writer. Core successor, rollback compatibility, publication,
+Project Successor, and project adoption remain separate authority boundaries.
+
+## Product outcome
+
+RunKit 0.18 is the Project Driver for work that continues across Agents,
+sessions, interruptions, and handoffs. A user should be able to return to a
+blank session, run `status` or `takeover`, and receive a bounded next action,
+the responsible Agent, the dominant gap, and the evidence needed to proceed.
+The normal recovery target is a useful takeover packet within two minutes on a
+local project with an existing definition and event history.
+
+## Normal typed mainline
+
+The ordinary user-facing flow is a typed direct CLI surface:
+
+1. `project init` accepts one project definition JSON file and creates the
+   durable project model.
+2. `project assign` gives a WorkItem to an Agent.
+3. `project checkpoint` records meaningful state and evidence.
+4. `project handoff` transfers a responsibility without overwriting its
+   history.
+5. `project decision` opens or resolves a decision with rationale and evidence.
+6. `project integrate` records an integration-gate result with evidence.
+7. `project status` projects the current team state.
+8. `project takeover` builds the next bounded action for a successor Agent.
+9. After the project derives `overall=completed`, `project successor`
+   archives it and installs one fresh project under explicit Owner invocation.
+
+Except for the one definition supplied to `init`, normal users provide typed
+flags and values. They do not need to author request, event, receipt, or hash
+JSON. The legacy `--request <event.json>` path remains available for existing
+automation and compatibility, but it is not the product's primary interaction.
+
+## Project Driver projection
+
+Every status and takeover projection exposes the same deterministic recovery
+spine:
+
+- `dominantGap`: the highest-priority unresolved decision, failed WorkItem,
+  blocked dependency root, active responsibility, or integration gate;
+- `nextAction`: the next bounded action that is legal from the durable state;
+- `nextActorId`: the Agent responsible for that action when one is known;
+- `headline`: a concise human explanation of the current project state.
+
+The projection derives these fields from append-only artifacts. It never accepts
+a guessed percentage, promotes an evidence candidate into business truth,
+invents completion, or overwrites an earlier assignment, handoff, decision, or
+checkpoint.
+
+## Continuity actions
+
+A handoff preserves the outgoing assignment and records the successor action.
+A decision separates the open question from its later resolution, rationale,
+and evidence. An integration event records that the project gate was reached;
+it is not a Git merge, release, deployment, or publication operation. Status and
+takeover are read-only projections and are safe to repeat after an interruption.
+
+## Bounded Project Successor
+
+`project successor` is the only 0.18 operation that replaces the active
+Project Driver truth. It first requires the current project to derive
+`overall=completed`; a chat summary, checkpoint claim, or caller flag cannot
+substitute for that projection. The target definition must use a different
+project identity and begins with no events.
+
+The operation holds an external lifecycle lock and the active project control
+lock, records a durable transition journal, archives the exact prior bytes of
+every regular file and directory under a content-addressed raw manifest, and atomically installs
+the fresh active project at `.owlcoda/runkit/project`. Exact retries resume
+the journal after the prepared, archived, or active-installed boundaries.
+Before completion, archived regular files and directories are sealed against
+ordinary writes with a receipt-bound `0444`/`0555` policy. An exact retry
+validates that archive's bytes and seal. Before any later successor writes a
+new journal or moves active truth, it validates the archive bound by every
+completed transition journal. Status and takeover remain active-truth
+projections and do not perform this historical scan. Different concurrent
+transitions, leaf or ancestor symlinks, non-regular files, archive collisions,
+truth drift, borrowed events, and ambiguous active state fail closed.
+
+This lifecycle operation does not select a new authoritative Agent, migrate an
+incomplete project, merge project histories, discover multiple projects, adopt
+a package, or grant repository or release authority. Its receipt records all
+Git, tag, publish, and deploy actions as false.
+
+## Deferred verification without test thrash
+
+RunKit may record a `verification_deferred` project event when related changes
+should settle before a focused or broad check is run. The event contains only
+stable check identifiers, the reason for deferral, the responsible Agent, the
+related WorkItem, and the exact integration gate before which it is due. It
+does not contain an executable shell command and does not schedule or run a
+test.
+
+An open record becomes the dominant project gap only after the ordinary work
+needed by that gate is complete. The declared gate cannot pass until the record
+closes. A `verified` close requires evidence; a `no_longer_required` close
+requires a previously resolved project decision. The second disposition is
+reported honestly and is never presented as a passing test.
+
+Coverage adoption, source-bound receipts, and `verify-plan` remain responsible
+for deciding which Formal commands are already covered and which are pending.
+The Project Driver consumes that outcome as coordination truth rather than
+duplicating the verification engine. This allows an Agent team to batch a
+small, explainable group of related changes and verify it once at the correct
+transition while preserving visible, fail-closed debt.
+
+## Hidden trust substrate
+
+The typed surface is backed by the existing trust substrate: immutable project
+definition identity, append-only and idempotent events, canonical bytes,
+regular workspace and path checks, source and evidence references, and
+execution-bound lifecycle truth where a WorkItem binds an execution. Config,
+profiles, leases, fingerprints, receipts, and the `.owlcoda/runkit` directory
+remain internal control data; they are not user-supplied authorization.
+The substrate continues to fail closed on malformed, ambiguous, redirected, or
+untrusted control state. Normal project coordination retains
+`authorizationGranted=false` unless a separately authorized boundary says
+otherwise.
+
+## Assurance boundary
+
+Multiple Agents coordinating a project do not make project coordination Formal
+by themselves. This normal flow does not make project coordination Formal by
+itself; multi-Agent coordination does not automatically mean Formal.
+The assurance lane is selected by the underlying risk and authority: ordinary
+coordination may remain on the typed Project Driver path; high-risk source
+integration, release, production, migration, destructive work, or an explicit
+Formal acceptance request crosses the assurance boundary. Quick and Formal
+verification remain distinct receipts and workflows. Neither grants Git, npm,
+deployment, production, funds, or business authority.
+
+## Performance and operating burden
+
+The Project Driver is local-first and daemon-free. A project needs one
+definition and an append-only event directory; ordinary users do not maintain
+lease, receipt, fingerprint, or request JSON by hand. `status` and `takeover`
+are read-only and deterministic, and repeated calls do not create project
+events. Projection cost grows with the bounded definition and event history,
+while the two-minute takeover target keeps interruption recovery operationally
+small enough for routine Agent changes.
+
+## Authority and compatibility
+
+`owlrunkit@0.18.0` is an independent standalone candidate. The root `owlcoda`
+package keeps its separate `0.15.32` lifecycle, and the published `owlrunkit`
+`0.17.2` package remains the trusted rollback prior. Existing 0.17.x project
+artifacts remain readable through the Core successor and trusted-prior paths.
+The 0.17 Agent-Native product definition remains bundled as historical
+compatibility documentation; this 0.18 document is the current Project Driver
+contract.

package/docs/architecture/OWLCODA_RUNKIT_USAGE_MODE_SELECTION_PRODUCT_DECISION_20260811.md

@@ -0,0 +1,142 @@
+# OwlCoda RunKit 使用模式选择：Sierac MES Dogfood 产品决议
+
+状态：**0.18.1 本地候选，未发布、未安装、未激活**
+日期：2026-08-11
+
+## 已验证现场事实
+
+以下结论来自对
+`/Users/yeemio/AI/project/sieracMes-AI/.owlcoda/runkit/` 的只读检查，
+不是体验推测：
+
+- W1.6 留下四份 Quick Receipt V1。按回执生成时的 `exitResult` 与
+  `mutationDecision` 分类，结果是 3 pass、1 fail、0 source mutation；四份均为
+  `source_unchanged`。
+- 旧版 `quick-metrics` 使用“当前 workspace 是否仍匹配历史回执”来统计历史结果。
+  在后续源码变化后，它把四份回执全部错算为 `sourceMutated`，并输出
+  `passed=0 / failed=0 / sourceMutated=4`。
+- Quick 的实际测试命令发现了 W1.6 的 pytest 问题；RunKit 的价值是保留失败、
+  绑定候选并阻止错误完成声明，不是替代项目测试发现业务缺陷。
+- 三次相关 Quick 的机器耗时约九秒。主要摩擦来自人工选择模式、填写标识和
+  时间戳、分别 checkpoint/integrate，以及排查不一致命令。
+- Project Driver 对长线程、跨会话恢复、脏工作树和延期验收有实际价值。
+- 现场成功回执长期携带 `quick_ignored_artifact_unbound`，但 attestation 仍可为
+  GO；该 issue 目前缺少清晰、可行动的严重级别。
+
+## 产品长期决策
+
+RunKit 定位为“飞行记录仪和阶段控制器”，不是所有命令的审批系统。
+
+用户看到五种模式，但内部不以单一枚举承担全部语义。路由始终保留两个正交维度：
+
+| 用户模式 | continuity | assurance | 典型用途 |
+| --- | --- | --- | --- |
+| `off` | `none` | `none` | 分析、文档、一次性探索、无需复用证据的小任务 |
+| `light` | `none` | `quick` | 单 writer、低风险、一个冻结候选、一条精确验证命令 |
+| `managed` | `project_driver` | WorkItem 级 `none/quick/formal` | 跨会话目标、交接、checkpoint、gate |
+| `formal` | 依连续性事实自动选择 | `formal` | 真正高风险或合同要求正式验收的边界 |
+| `auto` | 按事实选择 | 按事实选择 | 默认推荐 |
+
+模式只表达工作流偏好，不表达授权。任何模式都不得授予 Git、发布、部署、生产或
+业务权限。
+
+下列事实强制 Formal：多 writer、权限或安全边界、资金、不可逆变更、回滚不确定、
+schema/data migration、明确 Formal acceptance。`off/light/managed` 不能覆盖这些
+事实。
+
+下列标签或事实单独出现时不强制 Formal：普通项目源码修改、`longRunning`、
+`interruptionRecovery`、production、release、deployment。长周期和中断恢复主要
+选择 Project Driver；单 writer、可回滚且有业务读回的普通源码变更可以是
+`managed + quick`。风险被 containment 后必须重新路由，升级不具有粘性。
+
+## 0.18.0 已有能力
+
+- Quick Verification 保存 source-bound 命令回执、stdout/stderr 和失败状态。
+- Project Driver 保存目标、WorkItem、依赖、checkpoint、handoff、decision、
+  verification gate 和 takeover 状态。
+- Formal Delivery 提供候选、lease、严格证据与正式验收边界。
+- 所有这些能力都保持 `authorizationGranted=false`，验收与 Git/发布/部署/业务授权
+  分离。
+
+0.18.0 并没有提供五模式入口，也没有解决本次现场暴露的历史 metrics、Quick 后续
+命令、`quick-attest --help` 和 Skill/Core 版本漂移诊断。
+
+## 本轮 0.18.1 候选实现
+
+### 使用模式只读入口
+
+新增：
+
+```bash
+owlrunkit mode recommend --workspace "$PWD"
+owlrunkit mode status --workspace "$PWD"
+```
+
+普通输出为人类可读文本；`--json` 提供严格机器结果。推荐结果包含：推荐模式、
+continuity、assurance、触发原因、被拒绝的较轻模式、预计新增步骤和证据数、下一条
+命令、降级是否允许，以及完整权限边界。
+
+`mode status` 当前只报告 `default_auto`，不创建 `.owlcoda`，也不暗示已有持久化
+设置。本轮没有实现 `mode set`。
+
+### MVP 一致性修复
+
+- `quick-metrics` 以历史回执自身的 `exitResult` 与 `mutationDecision` 统计 pass、
+  fail 和 source mutation；当前 workspace 漂移不再改写历史结果。回执 schema、
+  snapshot binding 和输出材料仍校验；Core/context 的信任结论继续由
+  `quick-attest` 独立报告，不改写历史运行统计。
+- Quick 结果的后续命令改为可执行的
+  `owlrunkit quick-attest --workspace ... --receipt ...`。
+- `quick-attest --help` 返回 action-specific help。
+- assurance router 不再因普通 project mutation、long-running、interruption
+  recovery 或 production/release/deployment 标签自动升级 Formal；真正强制事实仍
+  fail closed，且解除后可以降级。
+- Skill installer 的只读 `inspect` 同时检查受管文件完整性与运行包的
+  Skill/config Core 版本；版本旧时明确返回 `version_mismatch`。
+- 同一候选内的 profile 一致性修复不再为不存在、也无法从真实脚本/workspace
+  推导出的 `src/**`、`tests/**` 制造阻塞项，并提供现有 profile 的原子 reconcile。
+
+## 后续缺口
+
+### P1：Quick warning 需要可行动语义
+
+把 `quick_ignored_artifact_unbound` 区分为不影响当前决策的 maintenance note 与会
+阻断复用的 issue。验收标准：GO 回执不再永久携带无法行动的高噪声告警；任何被
+降级的提示仍在详细输出中可查，真实材料缺口继续 fail closed。
+
+### P1：Project Driver 日常动作减负
+
+自动生成安全的事件 ID 和时间戳，并提供一个原子“完成 WorkItem 并在条件满足时
+integrate”动作；`completedUnits` 可从已声明的单工作项单位推导。验收标准：普通
+单项完成不需要手填 ID/时间戳，也不需要两次独立命令，同时 append-only 事件和
+重试幂等性不变。
+
+### P1：验证指纹按真实路径范围绑定
+
+Quick/coverage 应区分验证声明覆盖的源码、仅文档/空白变化和真正全局 gate。验收
+标准：不相关文档或空白变化不会诱导重复大测试；声明范围内变化、lockfile 变化和
+global gate 仍正确失效。
+
+### P2：持久化模式偏好
+
+只有在只读推荐经过更多 dogfood 后再评估
+`owlrunkit mode set --mode auto|off|light|managed|formal`。验收标准：偏好具有明确
+来源和审计读回，且永远不能覆盖强制 Formal 路由。
+
+## 不能声称的内容
+
+- 这份文档和当前 worktree 不是已发布的 0.18.1，也不是 npm、Core、Skill 或任何
+  项目的升级完成证明。
+- `mode recommend` 根据调用者提供的结构化事实路由；它不会自动发现所有业务、
+  权限、资金或迁移风险。
+- RunKit 降低进度、证据和完成状态幻觉，不直接解决 Sierac MES 业务问答幻觉。
+  后者仍由 canonical 接口、schema、adapter 与 Answer Harness 负责。
+- Quick 不是 Formal acceptance，Project Driver 不是测试执行器，任何 GO/accepted
+  结果都不授予 Git、发布、部署、生产或业务权限。
+
+## 版本建议
+
+本组变更适合形成 `owlrunkit@0.18.1` 小版本候选：它保持 0.18 的 Project Driver
+和回执合同，新增只读 CLI，并修复 dogfood 验证出的行为缺陷。候选只有在聚焦测试、
+RunKit contract/package gate、Core 身份同步、pack 内容检查和安装 smoke 都通过后
+才具备发布评审资格；发布、项目升级与运行时激活仍是之后的独立动作。