ultra-builder-pro-cli 0.26.2 → 5.0.0

package/docs/WORKFLOW-LIFECYCLE.md

@@ -1,14 +1,20 @@
 # Ultra Builder Pro workflow lifecycle
 
-Ultra v0.26 has no fixed runtime state machine. Skills define checkable entry and exit
+Ultra 5.0 has no fixed runtime state machine. Skills define checkable entry and exit
 criteria; repository files record facts; the host model chooses the route.
 
 ## Capability flow
 
 ```text
-ultra-init ──raw Project Brief──► ultra-research ──accepted baseline──► ultra-change
-                                      ▲                                  │
-                                      └──── scoped consequential evidence gap ────┘
+ultra-init/create|recover ──raw Project Brief──► ultra-research/baseline
+                                                       │
+                                                       ▼
+                                               ultra-change/intake
+                                                       │
+                   ┌── ultra-research/change-scoped ◄──┤ evidence-needed
+                   └── ultra-change/reconcile ─────────┘
+                                                       │ accepted
+                                                       ▼
 
 ultra-change ─► ultra-plan ─► ultra-dev (per task) ─► ultra-test ─► ultra-deliver
                      ▲                 │                    │
@@ -18,16 +24,47 @@ ultra-status    read-only router at any point
 ultra-delegate  orthogonal bounded execution on another CLI
 ```
 
-Arrows are recommendations, not automatic invocation. The owner explicitly selects
-each public workflow. Delivery archives one stable `change_id`; the next request opens
-a new active id while prior task rows remain in the append-only ledger.
-
-## Common entry contract
+Arrows are recommendations unless a live owner-activated Execution Grant authorizes the
+exact next mode. Public modes are either `explicit-entry`, `grant-continuable`, or
+read-only `router`; model-only methods are `model-discipline`. Scoped Research,
+same-Change reconciliation, Plan, Dev, Test, and Deliver reconciliation may continue
+only when the grant covers the same `change_id`, mode, effects, and stop conditions.
+Init, Change intake/evidence/acceptance/abandonment/primary transfer, Delegate, Deliver
+finalization, archive, and every external effect remain explicit-entry. Status remains
+owner-invoked, read-only, and supplies no continuation authority. Delivery archives one stable
+`change_id`; the next request opens a new active id while prior task rows remain in the
+append-only ledger.
+
+## Topology and checkpoints
+
+The owner chooses the Agent topology at every stage: one Agent or several, which
+providers, which write scopes, serial or parallel. The unspecified default is the
+current Agent continuing alone — no automatic spawning, delegation, or control-plane
+enablement, and no provider permanently bound to a role. One primary writer owns the
+canonical `.ultra` files in a worktree; parallel source writes use isolated worktrees
+with explicit integration. Moving the primary-writer role itself to another Agent is
+an owner-granted primary transfer — OFFER, verified ACK, sole-writer execution,
+frozen terminal RESULT — and never a widened delegation
+(`skills/ultra-change/references/primary-transfer.md`).
+
+Every owner-facing checkpoint — a WIP at a stop, a closing Resume Note, a delivery
+report — carries the eight fixed semantics defined in
+[artifact authority](ARTIFACT-AUTHORITY.md#owner-checkpoint-semantics): why, outcome,
+accepted boundary, delta, reality, decision needed, next bounded action, and not-done.
+The shared [Route Contract](../skills/ultra-status/references/route-contract.md) adds
+Route, Mode, Position, and Exit condition as navigation around those semantics. Every
+public Skill shows the card at entry, a material checkpoint, pause/blocked exit, normal
+exit, and before a public transition. Cards are derived observations, not stage markers,
+permission, or persisted route state; alignment failure stops work until the owner
+re-accepts an outcome.
+
+## Common mode-aware entry contract
 
 Every user-invoked Skill reads, in order:
 
 1. zero or one `.ultra/changes/active/<change_id>/intent.md`; more than one is a conflict;
-2. `.ultra/tasks.json`, selecting only tasks whose `change_id` matches that active id;
+2. `.ultra/tasks.json`, selecting only tasks whose `change_id` matches that active id
+   and treating its v2 row as the sole task-status authority;
 3. the matching frontier task's `context_file` and closing `## Resume Note`;
 4. `.ultra/project-brief.md`, the accepted `.ultra/north-star.md`, `CONTEXT.md`, and
    relevant `.ultra/decisions/`, where those artifacts already exist;
@@ -37,6 +74,11 @@ Every user-invoked Skill reads, in order:
 Missing files produce a precise recovery or initialization recommendation. They do not
 cause the Skill to invent a state transition.
 
+After reading these facts, the Skill chooses only one mode declared in its own Route
+table and emits the entry Route Card. If model-selected, it revalidates that this mode is
+grant-continuable and that the grant covers the same Change and effects. Host metadata
+is only coarse discovery; it never substitutes for this runtime check.
+
 ## Initialization
 
 Init writes the raw Project Brief and stable empty skeleton. `ultra-init` preserves the
@@ -48,22 +90,29 @@ launch another workflow.
 
 ## Research
 
-Research turns the brief into an accepted North Star and evidence-backed baseline.
-`ultra-research` uses Wayfinding only when a multi-lens path is unclear, then maps the
+Research has two modes. `baseline` explicitly turns the brief into an accepted North
+Star and evidence-backed baseline. `change-scoped` answers only one active Change's
+recorded Research Disposition and returns named evidence to that same Change for
+reconciliation. `ultra-research` uses Wayfinding only when a multi-lens path is unclear,
+then records entry question, selected lenses, omissions, dependencies, source plan,
+material checkpoints, observable stop condition, and return route before mapping the
 question to the smallest useful subset of seventeen focused references. The first six
 are semantic lenses with real dependencies: `00` precedes `01`; conditional `02` and
 `03` may fan out after the boundary is known; `04` consumes their relevant conclusions
 and is the first owner checkpoint; `05` extracts assumptions from `00` through `04`.
 Grilling controls how missing owner inputs are asked, Think tests one consequential
 decision, and Domain Modeling writes settled vocabulary; none owns the Research route.
-The final synthesis promotes accepted conclusions into the North Star, `CONTEXT.md`,
-three specifications, and a source-hash-bound distillate.
+The baseline synthesis promotes accepted conclusions into the North Star, `CONTEXT.md`,
+three specifications, and a source-hash-bound distillate. A change-scoped run updates
+only its mapped sections and never changes Change acceptance on its own. Lens counts are
+navigation facts, not semantic progress or a completion signal.
 
 ## Change reconciliation
 
-Change updates only the touched baseline rather than rebuilding it. `ultra-change`
-first compares requested behavior with accepted product truth and current code. It
-classifies changes by the resulting commitments:
+Change updates only the touched baseline rather than rebuilding it. Its visible modes
+are `intake`, `evidence-needed`, `reconcile`, and `accepted`. `ultra-change` first
+compares requested behavior with accepted product truth and current code. It classifies
+changes by the resulting commitments:
 
 - expansion and correction may proceed inside accepted intent;
 - a reduction waits for explicit owner authority.
@@ -74,12 +123,23 @@ existing evidence; `bounded` or `required` names the exact question, lenses, and
 evidence. Unsatisfied exit evidence routes to Research, then back through Change
 reconciliation before planning when the accepted contract needs an update.
 
+Draft intent is mutable. Revising a consequential accepted answer creates a superseding
+decision record so Status can reconstruct history without treating an event log as
+authority. Only same-Change, non-material `reconcile` may continue under a valid grant;
+opening or accepting a boundary, abandoning it, reducing a commitment, or changing the
+canonical writer is an explicit owner checkpoint.
+
+If the owner rejects the offered options as inadequate, Change preserves that answer,
+keeps the draft unresolved, and returns to evidence or materially different options. It
+never selects the closest rejected option or advances to Plan by default.
+
 A micro edit that makes no specification sentence false stays outside the Ultra
 lifecycle and uses ordinary repository TDD/verification. Once an active Change exists,
-even a `quick` profile receives one minimal Plan task so evidence and delivery remain
-bound to its id. A separate request cannot open a second active Change: finish or
-deliver the current one, abandon it with owner authority, or explicitly reconcile the
-new request into the same stable id.
+Plan creates the smallest graph justified by the real seams, evidence, review, and
+recovery needs so delivery remains bound to its id; a profile never fixes task count. A
+separate request cannot open a second active Change: finish or deliver the current one,
+abandon it with owner authority, or explicitly reconcile the new request into the same
+stable id.
 
 ## Planning
 
@@ -88,27 +148,81 @@ current-Change task graph and one context file per task without deleting histori
 rows. The model chooses ordinary technical seams; the owner is asked only when a seam
 changes a public contract, accepted risk, or another material trade-off. Every task
 stores the stable `change_id`; dependencies stay inside that Change and ids remain
-globally unique. Feature tasks are tracer bullets. A contract task precedes cross-process
-or cross-top-level-boundary implementations; integration checkpoints appear after
-several slices. Cycles and broken trace paths are repaired before handoff.
+globally unique. Feature tasks are tracer bullets. A contract task precedes dependent
+implementations when a shared contract must remain valid across independently changed
+boundaries; integration checkpoints appear where slices first compose or observed
+coupling and recovery risk justify a whole-path check. Cycles and broken trace paths are
+repaired before handoff. Complexity, file, task, line, and context counts are
+observations rather than plan-quality gates.
+
+Plan names the earliest executable tracer bullet's real consumer, real data path, and
+owner-visible observation. A product or contract evidence gap returns to
+`ultra-change/reconcile`; Change decides whether bounded Research is required. Ordinary
+technical uncertainty remains model-owned or becomes a bounded spike/tracer task, so
+Plan never bypasses Change by opening product Research directly.
 
 ## Task development
 
 `ultra-dev` writes the implementation plan into the task context before code. It uses
-`ultra-tdd` on confirmed seams, records six evidence dimensions, and keeps ledger and
-context status synchronized. The closing Resume Note is rewritten even when work stops
-early.
-
-Task review uses six independent lenses. Findings move through stable JSON artifacts,
-not intermediate worker chatter. Refactoring happens after the evidence makes real
-duplication or coupling visible.
+`ultra-tdd` on confirmed seams and records `ultra-task-evidence-v2`: typed acceptance,
+six evidence dimensions, subject/context identity, limitations, and task-review
+provenance. `.ultra/tasks.json` is the sole task-status authority; a new context has no
+Status field, while legacy Status/Complexity values are migration diagnostics and the
+ledger wins. The closing Resume Note is rewritten even when work stops early.
+Before publishing that record, Dev hashes bounded stable repository-contained bytes for
+every command or external `raw_evidence_ref` into `raw_evidence_sha256` and reads both
+the raw receipt and written record back. Structural validation checks ref/digest shape
+without dereferencing the receipt.
+
+Task review selects lenses by risk and touched seams while the ledger row remains `in_progress`.
+Its first admission binds the ledger, task context, immutable packet scope, and actual
+pre-review evidence. It does not require a final v2 record or claim a task worktree
+digest that Worker Packet v1 does not carry.
+Findings move through stable JSON artifacts, not intermediate worker chatter. Verdicts
+are terminal for the current subject: `APPROVE` ends the task review even when P2/P3
+findings are retained — they become report entries or owner-selected backlog, never a
+fresh review of the same subject. `REQUEST_CHANGES` routes only the exact current
+P0/P1 findings as one in-scope repair set followed by one affected-lens delta review
+per repair set inside the active review budget — an exact current owner grant
+overrides the versioned product default of one initial review plus at most two
+P0/P1 delta reviews per package; a `REQUEST_CHANGES` that would exceed the active
+budget is an owner checkpoint, not another automatic repair, and the same root
+surviving three failed fixes stops point-patching and reports an architecture
+problem. Lens selection follows the review kind, never a fixed count: an initial task
+review selects `review-spec` plus the lenses justified by risk and touched seams, a
+delta review reruns only the affected lenses, and an aggregate Change review may
+default to all six only when cross-task wiring justifies it. Budget exhaustion keeps the row `in_progress` and returns
+`owner checkpoint` / `budget_exhausted` without any semantic verdict.
+After the final summary validates, Dev independently captures the `subject`
+as a completion-snapshot freshness observation and immediately writes the one canonical
+v2 record; its `task_review` separately binds the retained strict summary, which does not
+prove the subject worktree digest. Only then may the primary model write `completed` to
+the ledger. Refactoring happens
+when evidence makes real duplication, coupling, or a boundary defect visible, never
+because a slice, finding, file, or repair-round count crossed a threshold.
+
+If later work invalidates completed evidence, Dev records the affected criterion IDs and reason in the context Change Log and Resume Note before changing the
+ledger-authoritative row from `completed` to `in_progress`; there is no silent demotion.
+The old frozen review remains historical evidence while the reopened task refreshes the
+affected observations, obtains a new final review, and captures a new completion subject.
 
 ## Whole-system audit
 
-`ultra-test` runs once after tasks for the active Change are complete. It checks anti-patterns, coverage,
+`ultra-test` runs once after tasks for the active Change are completed in the ledger
+with current v2 evidence and retained task-review provenance. Historical v1 evidence is
+a migration diagnostic, not current Test input. It checks anti-patterns, coverage,
 wiring, E2E behavior, performance, and security. It names the exact Git commit and
 binds the report to `change_id`, exact current task ids, intent SHA-256, and the
-product-worktree digest. `passed: true` implies complete current tasks and no required
+product-worktree digest for the current whole Change. Each task evidence subject remains
+its independently captured completion-snapshot freshness observation rather than being
+rebound to this aggregate identity. Test independently rechecks current Acceptance bytes,
+criterion IDs and verification types, retained task-review summary, durable owner record,
+and cited affected artifacts. For command and external evidence it independently takes
+the bounded stable non-symlink raw bytes, recomputes `raw_evidence_sha256`, then recomputes
+the containing record's `evidence_digest`. `.ultra/evidence/**` is excluded from the
+product-worktree digest, so these separate exact bindings prevent a self-referential
+receipt. An invalidated task returns to the explicit Dev reopen
+path. `passed: true` implies complete current tasks and no required
 area left `not_run`; task completion does not force a pass, and dispositioned gaps stay
 visible. It never turns an orphan,
 failed command, stub, or omission into permission to alter scope.
@@ -134,18 +248,32 @@ effects. Readiness grants none automatically.
 `ultra-status` infers the smallest next route from observable files:
 
 - no healthy skeleton or usable raw brief → init;
-- raw brief without an accepted baseline, or a consequential evidence gap → research;
+- raw brief without an accepted baseline → baseline research;
 - requested outcome without active Change → change;
 - draft or blocking active intent → change;
-- unsatisfied Research Disposition → research;
+- unsatisfied Research Disposition → change-scoped research, then the same Change reconciliation;
 - active Change without matching executable tasks → plan;
-- matching pending or in-progress task → dev;
-- completed current-Change tasks without a matching fresh report → test;
+- the unique matching `in_progress` task → dev;
+- only `pending` tasks remaining → report them as frontier candidates and recommend
+  one exact task id for an explicit owner invocation of dev;
+- a completed ledger row with missing, stale, or invalid v2 evidence/task review → dev
+  to record affected criteria and reason, explicitly reopen, and repair provenance;
+- completed current-Change tasks with current v2 evidence and no matching fresh report → test;
 - false report → route by finding and owner disposition, not by the boolean alone;
 - fresh, dispositioned report and unarchived Change → deliver.
 
-It also checks installation provenance, stale reports, old in-progress work, and
-unresolved markers. It does not mutate artifacts.
+It also checks installation provenance, unresolved markers, broken resume paths, and
+report freshness by root Change identity, exact ordered v2 task-evidence/review
+identities, and strict aggregate Review packet/admission/subject/summary bindings. It
+replays the existing structural validators; they observe transport and freshness but
+never decide semantic acceptance. Age, complexity, and file/task counts may be reported
+as context but never decide staleness or quality. It does not mutate artifacts.
+
+Status outputs the Route Card by default and expands `history`, `installation`,
+`delegation`, or `recovery` on request. An ambiguous or malformed active authority stops
+semantic routing and writes, but not diagnosis: recovery mode reports the typed fact,
+safe Git/directory observations, and the cheapest repair/retry/cancel/abandon path
+without selecting an ambiguous intent as truth.
 
 ## Delegation
 
@@ -156,6 +284,10 @@ the latter two may be read-only and do not require a task, so both pre-Plan evid
 post-task review remain reachable. Every scope still binds the one active `change_id`, a
 clean isolated worktree, strict permission/result schemas, and primary-host inspection.
 The worker cannot write canonical `.ultra` files or perform external effects.
+Delegation itself is explicit-entry. Tool calls, plans, usage, permissions, process
+output, ACP updates, and terminal transport state remain execution observations; only
+the primary writer interprets the validated result against Research, task, review, or
+Test acceptance and selects the semantic route.
 
 ## Artifact closure
 
@@ -170,14 +302,19 @@ Every durable output has a later reader:
   planning, development, audit, review, status, and delivery; an abandoned intent has
   an exact `## Abandonment` closure consumed by future Change reconciliation and Status;
 - the append-only task ledger and contexts are the common resume input after filtering
-  by active `change_id`, while task evidence feeds
-  review, whole-system audit, and delivery;
+  by active `change_id`, while typed v2 task evidence and retained task-review
+  provenance feed review, whole-system audit, and delivery;
 - the current test report is consumed by status and delivery;
 - delivery moves with the Change into archive and remains historical evidence.
 
 `ultra-status` intentionally writes nothing. Review, Hook, compact, and delegation
-files are derived observations with explicit rebuild or rerun paths. No route creates
-an unowned drift log, Change plan, technical-debt report, or second delivery summary.
+files are derived observations rather than semantic authority. A current strict review
+session contains `WORKER-PACKET.json`, `ADMISSION.json`, every selected specialist
+artifact, and `SUMMARY.json`; retain it until aggregate Test and Deliver have both consumed it
+successfully, then it may be garbage-collected. Premature loss requires a fresh Review
+and Test; never reconstruct the old receipt. Other derived observations keep their
+explicit rebuild or rerun paths. No route creates an unowned drift log, parallel planning file,
+technical-debt report, or second delivery summary.
 
 ## Interruption and cross-host continuation
 
@@ -188,11 +325,26 @@ task, evidence, or report authority would lose meaning.
 
 Before stopping, the active task context records current facts, exact verification,
 open questions, Completion, and a closing Resume Note. The next session or host repeats
-the common entry contract and compares the working tree with those claims.
+the common entry contract, compares the working tree with those claims, and derives a
+fresh Route Card. It does not rerun completed Research merely because conversation
+memory is gone. When context pressure threatens a reliable continuation, the current
+Agent finishes the nearest safe write, refreshes the Resume Note, shows a stop card, and
+recommends a fresh session; context size does not become route authority.
 
 This path remains valid when hooks and `ubp` are disabled. A compact snapshot may be
 used as a hint, then checked against canonical files and Git.
 
+Continuing on a different Agent — not merely a different host — additionally
+requires a primary transfer: an exact owner grant, a derived OFFER binding the
+canonical refs, HEAD, and worktree digest, the receiver's stable-read ACK, and a
+frozen terminal RESULT (`skills/ultra-change/references/primary-transfer.md`).
+Receipts are derived state; when they are lost, the owner confirms and a fresh
+handoff id is issued rather than reconstructing an old ACK. The post-review
+task closeout never refreshes that RESULT or opens a handoff for itself: beside
+a newest completed v2 RESULT it publishes the handoff's CLOSEOUT receipt — the
+one prescribed, uncommitted, review-and-handoff-free closeout transition of the
+same contract.
+
 An authenticated Claude-to-Codex continuation using this path passed on 2026-08-03;
 the exact scope, provenance, results, and limitations are recorded in the
 [runtime compatibility matrix](RUNTIME-COMPAT-MATRIX.md#authenticated-cross-host-continuation-evidence).

package/docs/evals/adversarial-review-2026-08-14.md

@@ -0,0 +1,84 @@
+# 对抗性审查盲测记录 — 2026-08-14
+
+## 结论
+
+本轮证据不支持增加永久第七个 `premise` lens。当前 `review-spec` 在明确要求校验
+North Star 因果贡献后，已经覆盖候选 lens 的五类已知缺陷；候选 lens 没有稳定找到
+当前 roster 漏掉的 consequential defect class。
+
+因此保留六 lens roster，把 premise challenge 固化到 `review-spec`，并把独立
+`premise` 角色保留为 Plan 与 aggregate Test 的 blind cross-family probe。这个 probe
+是隔离采样，不参与投票，也不能用无效结果替代六 lens findings。
+
+## 评估边界
+
+- Fixture revision: `37c07dc8596cda64a93aac260917b3c091ad1b58`
+- Hidden seeds: `tests/evals/adversarial-review-seeds.json`
+- Current prompt: `tests/evals/prompts/current-spec-lens.md`
+- Candidate prompt: `tests/evals/prompts/candidate-premise-lens.md`
+- Six-lens probe: `tests/evals/prompts/current-six-lens-probe.md`
+- Permission: read-only、无 external effects、独立 Git worktree
+- Claude Code: `2.1.232`
+- Kimi Code: `0.31.1`，显式模型 `kimi-code/k3`
+- Grok Build: `1.0.3 (1a29d5bc12d4)`
+- ZCode CLI: `0.16.3`，临时非持久化 provider 配置，GLM-5.3
+
+Seed manifest 从所有 Worker Packet 中隐藏。所有有效结果都通过
+`ultra-delegation-result-v1`、packet digest、只读 diff 和 launcher strict validation；
+模型陈述“未修改文件”本身不算权限证据。
+
+## Seed 映射
+
+| Run | Premise | Correctness | Test coverage | Recovery | Documentation | 可计分 |
+|---|---:|---:|---:|---:|---:|---:|
+| `spec-claude` / current spec | yes | yes | yes | yes | yes | yes |
+| `premise-claude` / candidate | yes | yes | yes | yes | yes | yes |
+| `six-kimi` / current six concerns | yes | yes | yes | yes | yes | yes |
+| `spec-zcode-3` / current spec | yes | yes | yes | yes | yes | yes |
+| `spec-grok-4` / current spec | partial raw text | partial raw text | partial | no stable result | partial | no |
+| `premise-grok-4` / candidate | partial raw text | partial raw text | partial | partial | partial | no |
+
+Claude 的 candidate 产生了比 current spec 更多的表述和风险拆分，但没有新增 seed class。
+Kimi 的完整六关注面与 ZCode 的 current spec 都再次找到五类 seed，说明 premise/causal
+挑战可以被最窄地放进 specification fidelity，而不必增加一个永久 lens。
+
+Grok 四轮尝试均以 exit code 0 结束，但输出被 host wrapper 截断或混入重复、畸形 JSON，
+launcher 正确发布 `failure_type: missing_result`。原始片段里出现若干真实缺陷不能算成功
+findings；这证明的是 Grok 1.0.3 的 structured-result conformance 缺口，不是 lens 语义结论。
+
+## 原始观察
+
+原始 receipt、worker spec、stdout、stderr 与 terminal result 保存在可重建观察区：
+
+`.ultra/.runtime/adversarial-eval/2026-08-14/delegations/`
+
+关键 terminal result SHA-256：
+
+- `spec-claude/result.json`: `14156c2fd86936a725fe5e9e42b90e8cb9e80e2d3045304046bd9791baa5caaf`
+- `premise-claude/result.json`: `bf39a38b5893f53a18f95f55932b28e381edbf2f4a8bae9b1dd1180b61267bbd`
+- `six-kimi/result.json`: `ca523134aa30f2ad39ebf539dd2b31b74b06c966c37d4cee4feae17574bf6aaf`
+- `spec-zcode-3/result.json`: `f9409cdd62a5e41a5b183cbed155e25e097ec9a27d0d0bfae0f9e74d87468956`
+- `spec-grok-4/result.json`: `0d33a9f5f4a80d75b0ebb908d30f6d12d9e26be1f3122b976dfd126fba8a34de`
+- `premise-grok-4/result.json`: `6e33fd9fc6a17313151a640717ad257efbfb780417044f8b9d92cc46acd128b5`
+
+该目录是 disposable evidence，不是项目语义权威；可由 tracked fixture、prompts 和当前
+delegate launcher 重建。
+
+## 产品决定
+
+1. 六 lens roster 不变。
+2. `review-spec` 必须同时验证 accepted behavior、North Star Trace、因果链、contradiction
+   disposition，以及“全部 acceptance 通过但 outcome 仍失败”的路径。
+3. Plan review 与 aggregate Test review 必须产出六 lens session；没有 findings artifact
+   不得声称阶段完成。
+4. 可用不同 model family 时追加一次 blind premise probe；不可用时记录缺失，不伪造独立性。
+5. `execution_mode` 和每个 worker 的 coverage refs 必须落盘。顺序角色扮演只能记录为
+   `sequential_fallback`，不能声称 isolated multi-agent。
+6. 只有 fresh blind fixture 反复证明 current roster 漏掉独立 consequential class 时，
+   才重新讨论第七 lens。
+
+## 最便宜复测
+
+先运行 `node --test tests/adversarial-review-eval.test.cjs`，确认 green-test trap 与五个
+hidden seed 仍存在；然后在新的隔离上下文中用相同 permission 分别运行 current spec 与
+candidate premise prompts。不得向 worker 暴露 seed manifest 或本报告。

package/docs/evals/ultra-5-0-journey-results.md

@@ -0,0 +1,62 @@
+# Ultra Builder Pro 5.0 journey 验收结果 — 2026-08-26
+
+## 结论
+
+14 条已接受 journey 都有可观察结果并全部为 `PASS`。2026-08-25 的初始候选曾把
+ZCode ACP promotion 记为诚实 `STOP`；2026-08-26 的 live activation 先复现并修复
+ZCode 0.16.5 runtime-preferences handshake，再通过国内 BigModel provider 完成真实
+receipt，因此该行已经由新证据取代。之后六 Host real user-HOME install/Doctor、
+Windows descendant cleanup 和 same-commit 三平台 CI 也分别完成；公共 registry 与
+GitHub Release 仍由最后的 release task 独立验证。
+
+这里的证据分为三类，不能互相冒充：
+
+- `source-contract`：验证 Skill 得到一致、可到达、可停止的导航合同，不断言任一模型的
+  实际表达质量；
+- `isolated-runtime`：运行真实 installer/Doctor/runner 代码，但只使用临时 config、fake
+  Codex registration 或 fake ACP/CLI target；
+- `live-provider`：六个 Host 分别完成 one bounded read-only ACP provider session；
+  receipt 只保留协议/result/进程/Git 事实和 digest，不保留 prompt 或 provider text。
+
+## Journey register
+
+| ID | Journey | Result | Observable success or stop | Executed evidence | Deliberate gap |
+|---|---|---|---|---|---|
+| J1 | Greenfield | PASS | `init/create` 停在 raw brief，`research/baseline` 以 selected coverage 与 stop condition 产出 accepted baseline，再推荐首个 Change | `tests/route-contract.test.cjs`; `docs/evals/ultra-5-0-route-journeys.md#j1--greenfield-baseline` | source-contract；六 Host live output 未跑 |
+| J2 | Existing project | PASS | 新请求先进入 Change；有界证据问题经 change-scoped Research 回到同一 `change_id`，accepted 后才进入 Plan | `tests/route-contract.test.cjs`; route journey J2 | source-contract；不把模型语义质量机械化 |
+| J3 | Micro edit | PASS | 不使 specification 失真的 edit 明确退出 Change lifecycle，使用 ordinary repository TDD/verification | `skills/ultra-change/SKILL.md`; `docs/WORKFLOW-LIFECYCLE.md#change-reconciliation` | source inspection；实际 edit 由调用项目验证 |
+| J4 | Owner revision | PASS | draft 原地修订；accepted consequential answer 以 `supersedes` record 保留旧 authority；拒绝现有选项不会默认进入 Plan | route journey J3；`skills/ultra-change/SKILL.md`; `skills/ultra-status/SKILL.md` | source-contract；owner judgment 不由 fixture 代替 |
+| J5 | Fresh session | PASS | 从 root/HEAD、唯一 active Change、matching ledger/context/Resume、baseline 与 Git 重建 Route Card，不重跑已完成 Research | `tests/ultra-5-journeys.test.cjs`; README；`docs/WORKFLOW-LIFECYCLE.md#interruption-and-cross-host-continuation` | 本地 artifact route；本轮未做六 Host live output 对比 |
+| J6 | Early value | PASS | Plan 的第一个 slice 必须穿过真实 consumer 与 data path，并命名 owner-visible observation | `tests/ultra-5-journeys.test.cjs`; `skills/ultra-plan/SKILL.md`; lifecycle | source-contract；真实产品 observation 由调用项目提供 |
+| J7 | Degraded state | PASS | 两个 active Change 产生 `active_change_ambiguous`；不读任一 intent body，不写 progress，并给出显式保留/archive/abandon/retry 路径 | `tests/ultra-5-journeys.test.cjs` 真实 SessionStart Hook scenario；Status/Review source | Review 在 ambiguity 时停在 packet write 前；不会制造 current scope |
+| J8 | Clean install delegation | PASS | isolated install 后从 loaded `ultra-delegate/scripts/delegate.cjs` 直接运行；不依赖 global `ubp`，Doctor 可见 runner；六个真实 Host home 的最终 Doctor 也都选择 ACP | `tests/install.test.cjs`; `tests/package-smoke.test.cjs`; migration fixture；`.ultra/evidence/v50-real-home-activation/` | published-registry consumer install 由 public-release task 验证 |
+| J9 | ACP happy path | PASS | stable v1 initialize -> session/new -> bounded derived updates -> exact Ultra terminal result | `tests/acp-driver.test.cjs`; `tests/delegate.test.cjs`; fake ACP agent | synthetic executable protocol evidence，不是 Host/provider conformance |
+| J10 | Permission denial | PASS | envelope 内写请求仅 `allow_once`；scope widening 立即拒绝，不能转成 successful Ultra result | `tests/acp-driver.test.cjs`; `tests/delegate.test.cjs` | fake agent；没有真实 provider permission UI |
+| J11 | Cancellation | PASS | cancel/timeout 到 typed terminal result，释放 delegation lock，POSIX process group 无 descendant orphan；`windows-latest` 的 cancel 与 timeout 也都终止真实 grandchild，可用 fresh id 重试 | `tests/delegate.test.cjs`; `tests/acp-driver.test.cjs`; GitHub Actions run `32935138736` | 若 Windows `taskkill` 不可用，fallback 只能保证 direct child cleanup |
+| J12 | CLI fallback | PASS | 六个 Host 都只在自己的 current receipt、stable ACP v1 和 executable 同时成立时选择 ACP；任一条件缺失仍选择 result-compatible CliDriver，并在 receipt/result/Doctor 显示 typed reason | `tests/acp-driver.test.cjs`; `tests/install.test.cjs`; six live receipts；migration Doctor scenario | provider retry 与 fallback 不自动改变 credential 或语义 authority |
+| J13 | ZCode bridge | PASS | installed ZCode 0.16.5 通过 `zcode-acp` v0.1.0 + digest-bound compatibility patch 接入 ACP v1，并用国内 BigModel provider 返回 exact Ultra result、clean Git 与 terminal worker | `skills/ultra-delegate/patches/zcode-acp-0.1.0-zcode-0.16.5.patch`; bridge receipt；`host-zcode.json`; live conformance report | transport maturity 仍为 `experimental`，因为 `app-server` 没有公开稳定性合同；缺 bridge/executable 时仍 typed fallback |
+| J14 | 3.x migration | PASS | 18 个 seeded project-owned files（覆盖 baseline、spec、active/archive Change、decision、task/context、Research、evidence、test report 与 binary bytes）在 install -> update -> Doctor -> uninstall 前后 mode、length 与 SHA-256 全相同；六 Host Doctor healthy 且显示 driver health | `tests/ultra-5-journeys.test.cjs` before/after digest 与 Doctor diagnostic | fixture 证明 updater 的真实文件边界，不声称覆盖未知第三方 artifact 语义 |
+
+## 可复现命令
+
+```bash
+node --test tests/ultra-5-journeys.test.cjs \
+  tests/route-contract.test.cjs \
+  tests/acp-driver.test.cjs \
+  tests/delegate.test.cjs \
+  tests/install.test.cjs \
+  tests/package-smoke.test.cjs
+```
+
+其中 migration case 会在临时目录执行六 Host install、第二次 install（update）、JSON
+Doctor 与 uninstall，并在 TAP diagnostic 中打印 canonical before/update/uninstall digest
+和每个 Host 的 selected driver/fallback reason。
+
+## 尚待 public-release task 完成的验证
+
+- 六个真实 Host 对同一 route fixture 的 Route Card 输出质量对比；
+- `v5.0.0` annotated tag、npm trusted publication、GitHub Release 与 published-package
+  consumer install 的同源性。
+
+六个 live ACP provider receipt、real-HOME activation、Windows cleanup 和三平台 CI
+已经完成；剩余 registry/release 缺口属于 public-release task，不会被本报告冒充完成。

package/docs/evals/ultra-5-0-live-acp-conformance.md

@@ -0,0 +1,93 @@
+# Ultra Builder Pro 5.0 live ACP conformance — 2026-08-26
+
+## 结论
+
+本报告把每个 Host 的 ACP 激活结论绑定到自己的真实 provider receipt；一个 Host 的
+结果不能提升另一个 Host。Claude Code、Codex、OpenCode、Kimi Code、Grok Build 与
+ZCode 当前全部为 `PASS`，六个 profile 都可以在对应 executable 存在时选择 ACP。
+
+`PASS` 的共同条件是：stable ACP v1、真实 provider、精确
+`ultra-delegation-result-v1`、只读 isolated worktree、Git unchanged，以及 worker
+进入 terminal state。`end_turn` 本身不是完成证据；它必须和 exact result/diff
+validator 一起成立。
+
+## 方法
+
+命令：
+
+```bash
+node skills/ultra-delegate/scripts/acp-conformance.cjs \
+  --host <host> \
+  --output .ultra/evidence/v50-live-acp-activation/host-<host>.json \
+  --timeout 240
+```
+
+Harness 经过真实 `delegate.cjs`、background worker、`AcpDriver`、exact result
+validator 和 Git-diff validator。Receipt 只保存候选命令、协议/result 状态、事件数量、
+子进程终态及临时 artifact digest；不保存 prompt、provider text、hidden reasoning、
+credential value 或 raw JSON-RPC。临时工作树和 raw artifacts 在计算 digest 后删除。
+
+## Host register
+
+| Host | Candidate | Result | Receipt SHA-256 | Observation |
+|---|---|---|---|---|
+| Claude Code | `npx --yes @agentclientprotocol/claude-agent-acp@0.70.0` | PASS | `9b1d88cbd5e5b66a56e0cf19f34e7f2107f9e4b8fde9d80b9eadcdfc8f4a25e3` | ACP v1, `end_turn`, exact result, clean Git, worker exited |
+| Codex | `npx --yes @agentclientprotocol/codex-acp@1.6.2` | PASS | `e82287609397807fb6968dff104d7f6569499b53c77be6a186aa33939a4c629e` | ACP v1, `end_turn`, exact result, clean Git; adapter was terminated after the logical turn |
+| OpenCode | `opencode acp` | PASS | `3575b148f6feb8c9f1472add705390e6d47b87be080403ec7006b66a29f7873b` | ACP v1, `end_turn`, exact result, clean Git, worker exited |
+| Grok Build | `grok agent stdio` | PASS | `370eeae12d63443188d519825ed8040fd8bba9b27fddb82e499d9cb590f4a33d` | ACP v1, exact result, clean Git; creation-time metadata was bounded and identity-checked before use |
+| Kimi Code | `kimi acp` | PASS | `30d8adec2877f80fadf0b20c3cb7fa6b732dea3598d7484c143f9835770301fe` | ACP v1, `end_turn`, exact result, clean Git, worker exited |
+| ZCode | `zcode-acp --zcode-path <installed-runtime> --no-browser` | PASS | `156080f88ed84a013fb26d93e5138230732d2ef0e2b5b10a7c3deb501d3548e9` | domestic BigModel provider, ACP v1, `end_turn`, exact result, clean Git, worker exited |
+
+## Claude Code
+
+The pinned official adapter completed the real read-only drill. Its profile may select
+ACP when `npx` is executable; `ULTRA_DELEGATE_FORCE_CLI=1` remains recovery.
+
+## Codex
+
+The pinned official adapter completed the real read-only drill. A nonzero adapter exit
+caused by client-owned stdin shutdown remains recorded separately and does not overturn
+an already validated protocol terminal and exact result.
+
+## OpenCode
+
+The native stdio server completed the real read-only drill through `opencode acp`.
+
+## Grok Build
+
+Grok emits bounded creation-time metadata before returning `session/new`. Ultra accepts
+only four metadata update kinds at that point, caps the transient buffer at 32 messages
+and 256 KiB, and validates each buffered session ID after the response. Early content,
+tool traffic, mismatched identity, or overflow remains a typed failure.
+
+## Kimi Code
+
+Kimi 1.49.0 completed an owner-authenticated, read-only ACP provider session and returned
+the exact Ultra result. Its profile may select native `kimi acp` when executable.
+
+## ZCode
+
+Ultra resolves independently installed `zcode-acp` v0.1.0 source commit
+`42fe149d4b501469343c01f23ba3801832306d53`, the digest-bound Ultra compatibility patch
+`174dfd0feafe84320556a84b5ae08b10e44c5515fea4992fd1d58af54cee4e3d`, and installed
+ZCode CLI 0.16.5. The patch answers the new `session/requestRuntimePreferences` request
+that otherwise deadlocks upstream v0.1.0 during `session/create`. Locked Rust tests,
+the 0.16.5 regression, patched fake app-server E2E, release build, installed binary
+digest, exact bridge build marker, and the real provider session all pass.
+
+Authentication uses the Owner's existing domestic BigModel Coding Plan provider at
+`https://open.bigmodel.cn/api/anthropic` with model `GLM-5.2`; it does not use the
+international Z.AI OAuth flow. Credential values are absent from this report and its
+receipt. The transport remains `experimental` because the downstream ZCode app-server
+is not a public stability contract.
+
+## Recovery
+
+- Missing adapter, bridge, candidate identity, protocol version, or evidence returns a
+  typed CLI fallback reason before provider execution. If `session/new` reports
+  `acp_auth_required` before a prompt is sent, Ultra invokes the bounded CLI fallback
+  once; it never retries any other ACP failure.
+- `ULTRA_DELEGATE_FORCE_CLI=1` selects the bounded CLI path explicitly.
+- A failed or stale Host receipt is replaced only by a fresh bounded drill for that Host.
+- Raw receipt bytes remain repository evidence under `.ultra/evidence/`; this packaged
+  report carries their SHA-256 bindings without packaging private provider output.