@loomhelm/multi-task-flow 0.0.0-stage → 0.2.1

package/toolkit/skills/docs/vendor/dot-codex/specs/w7yqp-git-skill-convergence/SPEC.md

@@ -0,0 +1,176 @@
+# Git skill convergence
+
+> 当前有效规范以本文为准；实现覆盖与当前状态见 `./IMPLEMENTATION.md`，关键演进原因见 `./HISTORY.md`。
+
+## 背景 / 问题陈述
+
+- 本仓曾同时存在 git/local repo prompts、旧 git skills 与 PR/flow skills 中内嵌的 branch/worktree/rebase/cleanup 细节。
+- 这些入口让同一类本地 git 操作存在多份合同，容易在 prompt 清退、flow 恢复、PR 合并收尾时产生不一致。
+- 需要把纯 `git/local repo` 能力收敛成稳定 skill 面，让 PR/flow/design 入口只承担 orchestration 职责。
+
+## 目标 / 非目标
+
+### Goals
+
+- 固定新的 canonical git skill 面：`git-branch`、`git-worktree`、`git-sync`、`git-cleanup`，并保留 `git-remote-troubleshoot`。
+- 让 `pr-create`、`pr-merge`、`flow-continue`、`normal-flow`、`fast-flow`、`local-merge-flow` 与共享 workflow 只引用 git skill 名称，不再复制通用 branch/worktree/rebase/cleanup 细节。
+- 删除已退场的 git/PR/flow prompts 与旧 git skills，不提供 stub、alias 或 redirect。
+- 将 `work-branch-and-design` 保留为纯设计入口，不再创建分支、stash 或跳转旧 prompt。
+
+### Non-goals
+
+- 不批量迁移 `docs/specs/**`、replay、fixtures 中的历史名称。
+- 不改变 GitHub/PR convergence 业务语义。
+- 不改变 remote 通信方案、认证路径、upstream 或 push 目标。
+
+## 范围（Scope）
+
+### In scope
+
+- `skills/git-branch/**`
+- `skills/git-worktree/**`
+- `skills/git-sync/**`
+- `skills/git-cleanup/**`
+- `skills/pr-create/SKILL.md`
+- `skills/pr-merge/SKILL.md`
+- `skills/flow-continue/SKILL.md`
+- `skills/project-engineering-workflow/SKILL.md`
+- `skills/normal-flow/SKILL.md`
+- `skills/fast-flow/SKILL.md`
+- `skills/local-merge-flow/SKILL.md`
+- `skills/global-port-manager/SKILL.md`
+- `skills/docs-migrate/SKILL.md`
+- `skills/README.md`
+- `AGENTS.md`
+- `prompts/work-branch-and-design.md`
+- Deleted prompt files explicitly listed in this spec's acceptance criteria.
+
+### Out of scope
+
+- `docs/specs/**/replay/**`
+- dev-plan replay fixtures
+- historical implementation logs
+- unrelated prompts and skills
+
+## 需求（Requirements）
+
+### MUST
+
+- The only pure git/local repo skills are `git-branch`, `git-worktree`, `git-sync`, `git-cleanup`, and the existing `git-remote-troubleshoot`.
+- `create-worktree`, `remove-worktree`, `update-baseline`, and `pr-post-merge-cleanup` must be removed.
+- `goto-new-branch`, `goto-impl-branch`, `worktree-workspace`, `repo-cleanup`, `pr-create`, `pr-merge`, `pr-merged`, and `advance-flow` prompts must be removed.
+- `work-branch-and-design` must not contain `git switch`, `git stash`, old `/prompts:*` implementation handoffs, or a “create branch first” contract.
+- `git-cleanup` owns post-merge PR local closeout and local merge cleanup proof generation.
+- `git-sync` owns confirmed-base local sync, including direct detached `HEAD` alignment for clean maintenance worktrees, and replaces the old update-baseline role.
+- `git-worktree` owns manually managed worktree create/remove lifecycle and port lease release integration; native Codex Ticket Worktrees are created and archived only by native task lifecycle.
+- `git-branch` owns local branch create/switch/delete contracts.
+
+### SHOULD
+
+- Orchestration skills should describe when they call the git skills, not re-specify every low-level branch/worktree command.
+- Active reference checks should ignore historical `docs/specs/**`, replay, and fixtures unless they affect runtime entry behavior.
+
+### COULD
+
+- Future replay checkers may add focused cases for the new git skill routing surface.
+
+## 功能与行为规格（Functional/Behavior Spec）
+
+### Core flows
+
+- PR merge with `merge+cleanup` calls `$git-cleanup` after `$pr-merge` finishes merge-scope work.
+- Local merge cleanup calls `$git-cleanup MODE=local-merge` after `$local-merge-flow` completes the local merge and cleanup is explicitly requested.
+- Branch setup tasks route to `$git-branch`.
+- Worktree lifecycle tasks route to `$git-worktree`.
+- `$wave-thread-flow` Ticket parallelism routes to native Codex task Worktrees; it must not fall back to `$git-worktree ACTION=create`, and `$git-cleanup` must not remove native task internals.
+- Confirmed local base synchronization routes to `$git-sync`.
+- A clean detached maintenance worktree may use `$git-sync ACTION=align-detached-base` to fetch a confirmed base and remain detached at its exact commit. A bare mainline-sync request rebases only when `HEAD` is attached to a topic branch; shared topic-branch `BASE_REF` + `REMOTE` rebase calls keep their existing meaning.
+- Remote/auth/transport troubleshooting stays in `$git-remote-troubleshoot`.
+
+### Edge cases / errors
+
+- If a deleted prompt or old git skill name remains in active runtime docs, the migration is incomplete.
+- If a historical reference remains only in `docs/specs/**`, replay, or fixtures, it is not a blocker for this spec.
+- If cleanup cannot prove completion with live facts, it must report partial cleanup rather than hand-written success text.
+- Detached alignment must stop for dirty worktrees, ambiguous bases, unavailable old-commit recovery, or explicit rebase/fast-forward actions; it must not stash, create a backup branch, or alter remote configuration.
+
+## 接口契约（Interfaces & Contracts）
+
+### 接口清单（Inventory）
+
+| 接口（Name） | 类型（Kind） | 范围（Scope） | 变更（Change） | 契约文档（Contract Doc） | 负责人（Owner） | 使用方（Consumers） | 备注（Notes） |
+| - | - | - | - | - | - | - | - |
+| `git-branch` | skill | internal | New | None | main-agent | flow/design/git tasks | local branch operations |
+| `git-worktree` | skill | internal | New | None | main-agent | flow/parallel/dev tasks | worktree lifecycle |
+| `git-sync` | skill | internal | New | None | main-agent | flow/PR/maintenance tasks | confirmed local sync and detached alignment |
+| `git-cleanup` | skill | internal | New | None | main-agent | PR/local-merge cleanup | local closeout and proof |
+
+### 契约文档（按 Kind 拆分）
+
+- None
+
+## 验收标准（Acceptance Criteria）
+
+- Given the active runtime surface is searched outside `docs/specs/**`, replay, and fixtures When references are checked for deleted prompt names and old git skill names Then none of the deleted names remain as recommended runtime entries.
+
+- Given the skills directory is listed When git-themed skills are inspected Then `git-branch`, `git-worktree`, `git-sync`, `git-cleanup`, and `git-remote-troubleshoot` exist, while old git skills do not.
+
+- Given `work-branch-and-design` is inspected When its body is searched Then it contains only design-stage duties and no branch/stash/old prompt handoff contract.
+
+- Given cleanup scripts are tested When the closeout tests run from `skills/git-cleanup/scripts` Then post-merge closeout gate behavior remains covered.
+
+- Given a path is declared or detected as a native Codex task Worktree When cleanup context or gate evaluates it Then it is excluded from manual removal and returns the native archive handoff instead of `git worktree remove`.
+
+- Given a clean detached worktree and a confirmed remote base When `ACTION=align-detached-base` is used Then it fetches and switches to the exact base commit while remaining detached, without creating local backup branches or changing remote configuration.
+
+- Given a dirty detached worktree, or an explicit detached rebase/fast-forward request When `$git-sync` is invoked Then it stops before moving `HEAD`.
+
+- Given a detached commit that is not retained by the target or another ref and is absent from the HEAD reflog When `ACTION=align-detached-base` is used Then it stops before moving `HEAD` because no owner-authorized recovery path exists.
+
+## 验收清单（Acceptance checklist）
+
+- [x] 核心路径的长期行为已被明确描述。
+- [x] 关键边界/错误场景已被覆盖。
+- [x] 涉及的接口/契约已写清楚或明确为 `None`。
+- [x] 相关验收条件已经可以用于实现与 review 对齐。
+
+## 非功能性验收 / 质量门槛（Quality Gates）
+
+### Testing
+
+- Unit tests: `python3 skills/git-cleanup/scripts/test_closeout_context.py`
+- Unit tests: `python3 skills/git-cleanup/scripts/test_closeout_gate.py`
+- Unit tests: `python3 skills/git-cleanup/scripts/test_local_merge_closeout_context.py`
+- Unit tests: `python3 skills/git-cleanup/scripts/test_local_merge_closeout_gate.py`
+- Integration tests: `python3 skills/git-sync/scripts/test_detached_alignment_contract.py`
+
+### UI / Storybook (if applicable)
+
+- Not applicable.
+
+### Quality checks
+
+- `git diff --check`
+- Active reference grep for deleted prompt and old git skill names.
+
+## Visual Evidence
+
+PR: none
+
+- Not applicable: this spec changes text contracts and local scripts only.
+
+## Related PRs
+
+- None
+
+## 风险 / 开放问题 / 假设（Risks, Open Questions, Assumptions）
+
+- 风险：历史 specs/replay 继续保留旧名称，搜索时需要明确区分历史材料与活跃入口。
+- 假设：GitHub PR 业务语义不随本地 git skill 收敛改变。
+
+## 参考（References）
+
+- `skills/git-branch/SKILL.md`
+- `skills/git-worktree/SKILL.md`
+- `skills/git-sync/SKILL.md`
+- `skills/git-cleanup/SKILL.md`

package/toolkit/skills/docs/vendor/dot-codex/specs/wave-thread-flow/HISTORY.md

@@ -0,0 +1,37 @@
+# Wave Thread Flow History
+
+Wave Thread Flow was the historical successor to the former Initiative and Parallel direct PR coordination models. Their specifications, ADRs, and replay fixtures remain readable for historical audit, but no new plan may select those routes.
+
+The replacement preserves shared direct PR validation, review, publication, and release contracts while changing only the multi-task coordination boundary to a parent conversation, sibling child PRs, active event messages, and conversation-local parent state.
+
+## Four-Level Delivery Batches
+
+The historical model was Plan -> Delivery Batch -> Development Wave -> Task. A Delivery Batch was the unit of one final main-branch merge. One parent-child batch owned one parent PR and could schedule multiple waves; multiple parent PRs required multiple batches and an explicit reason for separate main-branch merges. A direct batch owned one wave and one task, whose implementation PR was also the final PR.
+
+Runtime state uses schema version 2 and keeps batch-local SHA lineage, integration progress, delivery milestones, and resource closeout separate. v2 task and batch events are idempotent by stable identity, reject payload conflicts, defer missing prerequisites, and support deterministic replay. Legacy v1 state remains readable through a `default` batch adapter without rewriting the source.
+
+Acceptance evidence is now bound to a task or batch scope and a lifecycle phase (`pre_merge`, `post_merge`, or `post_deploy`). Planned facts cannot prove publication or observation, and batch publication/deployment/observation must remain verifiably bound to the merged SHA lineage. Each new v2 batch carries an absolute observation_window.deadline; recovery cannot reset it. Conditions that are false are recorded as skips without resource creation; unknown conditions remain pending.
+
+The v2 readiness gates now bind each acceptance item to its owning task or batch, require non-empty evidence and assertions for every acceptance item, require task `pre_merge` evidence before `pr_ready`, require batch `pre_merge` evidence before final readiness, require batch `post_merge` evidence before `merged` or `published`, require batch `post_deploy` evidence before `deployed` or `observed`, and retain timeout facts when an observation deadline expires. False condition results now require and persist the exact plan approval receipt, new delivery SHAs require structured native provider lineage receipts, milestone collections reject duplicate or unidentified records without being rewritten during validation, failed observations become blocked timeout facts, finalization is transactional across batches while persisting a timeout observed during a failed attempt, and legacy v1 string lineage evidence remains readable only through the explicit compatibility adapter.
+
+Recovery now rejects self-authored completed merge payloads, parent-side merges without task pre-merge evidence, completed task evidence outside the current batch lineage, stale direct-PR base SHAs at Step 5C, passed milestones without bound prerequisite SHAs, late passed delivery milestones after timeout, and incomplete batch conditions or dependencies at Step 5C. v2 parent drift is recorded per batch, batch event scopes reject task/PR identity fields, and legacy events continue through an explicit adapter. Nested v2 tasks use the same native archive, closeout-target binding, and independent cleanup receipt APIs as legacy tasks, while each batch advances its own local wave without a cross-batch wave barrier.
+
+The v2 plan parser now requires explicit milestone-qualified task and batch dependencies, preserves case-sensitive owned-surface identity during orchestration/contract cross-checks, and applies the same lexical path normalization used by runtime overlap isolation.
+
+Runtime validation now applies the same explicit dependency and repository-relative surface rules to v2 manifests, treats required skipped acceptance as incomplete, rejects late publication after timeout, and preserves v1 multi-wave closeout when an earlier task merge remains on the legacy SHA history.
+
+The plan-contract entrypoint now treats nested v2 batch -> wave -> task orchestration as the canonical raw-plan contract and keeps the old duplicated Wave Thread Contracts format only behind the explicit historical JSONL adapter, while preserving case-sensitive target branch identities.
+
+The v2 shell route now recognizes standard batch topology labels after the batch identifier, and persisted passed milestones are rechecked against their required acceptance phase before recovery or finalization.
+
+Persisted `pr_ready` and `completed` tasks are now rechecked for task-level `pre_merge` acceptance during v2 validation and final readiness, closing the parent-child recovery path that could otherwise trust a downgraded task acceptance record.
+
+The v2 plan checker now requires every canonical batch's absolute observation deadline and every task's explicit `$skill` binding in Task Orchestration. Historical JSONL keeps its repeated section only for read-only compatibility; legacy string acceptance facts are read-adapted as `unknown` evidence while the original values remain preserved in `legacy_compat`.
+
+## Child Conversation Closeout
+
+The parent now treats `completed` as business completion only. After verifying that a child task has stopped and has no follow-up, it performs native conversation archive with an explicit envelope authorization and records a thread-bound, operation-bound receipt that remains pending on failure and is safe to retry. Worktree archive, branch cleanup, and port cleanup remain separate facts, so a merged child cannot be reported as fully closed until every applicable receipt is confirmed.
+
+## 移除
+
+按主人明确要求，删除 Skill 及底层历史兼容能力；不保留状态读取、事件回放或自动迁移。历史 ADR 与迁移基线作为证据保留。

package/toolkit/skills/docs/vendor/dot-codex/specs/wave-thread-flow/IMPLEMENTATION.md

@@ -0,0 +1,5 @@
+# 实现说明
+
+Wave Thread Skill、旧状态机、CLI、派发 Profile 回放和专属测试均已删除。共享编排与验收解析保留给 `multi-task-flow`。Python 迁移基线中的相关映射标为 `retired`，不再执行。运行时 CODEX_HOME 不在本变更部署范围内。
+
+现行多任务编排覆盖正向解析、Profile、Skill、依赖、归属、验收范围、拓扑和投影。迁移证据退休状态必须绑定原始基线来源；CI 不再调度已删除的 waves suite。其他直接流程的历史 JSONL 校验不属于 Wave Thread 状态兼容能力，继续保留。

package/toolkit/skills/docs/vendor/dot-codex/specs/wave-thread-flow/SPEC.md

@@ -0,0 +1,20 @@
+# Wave Thread 移除边界
+
+## Context and Scope
+
+Wave Thread 的 Skill、状态机、历史读取、事件回放和 CLI 不属于可用功能。多任务协调使用 `multi-task-flow`；现行编排解析器保留其共享职责。
+
+## Requirements
+
+- REQ-WTF-REMOVAL: 仓库不得提供 Wave Thread Skill、历史状态读取、回放、派发或迁移入口。
+- REQ-WTF-SHARED: 移除旧实现必须保留现行 `multi-task-flow` 的编排、验收及直接交付能力。
+- REQ-WTF-HISTORY: 历史 ADR、主题历史和 Python 迁移基线仅作证据保存，不得充当可执行入口；已退休映射不得宣称当前可执行验证成功。
+
+## Verification
+
+- VER-WTF-REMOVAL covers: REQ-WTF-REMOVAL, REQ-WTF-HISTORY. 验证 Skill 和 CLI 缺席，旧分派失败，迁移映射显式退休。
+- VER-WTF-SHARED covers: REQ-WTF-SHARED. 验证现行多任务编排、验收、Profile 和直接流程测试通过。
+
+## Related ADRs
+
+None

package/toolkit/skills/docs/vendor/dot-codex/specs/weppn-topic-spec-model/HISTORY.md

@@ -0,0 +1,31 @@
+# 主题级 Spec 模型与历史 specs hygiene 演进历史
+
+> 这里记录会影响 Agent 理解“为什么一步步变成现在这样”的关键演进；单次任务流水账不放这里，规范正文仍以 `./SPEC.md` 为准。
+
+## Decision Trace
+
+- `b39a914` docs(specs): adopt topic-level spec model
+- `b21cbfe` feat(skills): add solution sync and spec sync
+- `6ecce09` fix(spec-sync): honor legacy spec refs
+
+## Current Model Corrections
+
+- 初版 topic-level 模型纠正了“一任务一 spec”的问题，但把部分 `Status` / `Milestones` / `Change log` 细节删得过头了。
+- 当前版本改为：`SPEC.md` 只保留规范正文；实现状态与 rollout 事实迁到 `IMPLEMENTATION.md`；关键演进原因迁到 `HISTORY.md`。
+- 这样 Agent 仍然可以在 spec 目录内直接回答“现在做到哪了”“为什么会变成这样”，而不用回头跨 PR / commit 历史搜证。
+- 随后的 review 收敛又补了一层 legacy 兼容：在 canonical `docs/specs/**` 尚未建立前，legacy `docs/plan/**/PLAN.md` 仍必须被 `spec-sync` 视为有效同步目标；一旦通过 `$docs-migrate MODE=specs` 建立 canonical 目录，该目录也必须一次补齐 companion docs，避免刚迁完就被新的 drift gate 反咬。
+- 同一轮 review 还收紧了兼容边界：历史上已经 canonicalize、但仍是 `SPEC.md`-only 的旧 spec 目录不应被立刻硬阻断；只有当目录里已经开始使用 companion docs，才要求 `IMPLEMENTATION.md` / `HISTORY.md` 保持完整。
+- 后续将新建身份从 ID 改为稳定 topic slug：历史 ID 目录不批量重命名，读取与 Git rename 映射继续兼容；新建目录、related_specs、视觉证据与 PR metadata 一律使用 `docs/specs/<topic>/`。
+- Legacy drift mapping now requires proof before stripping an apparent ID prefix: an exact historical catalog `ID + source path` row or a same-diff Git rename. This prevents valid slugs such as `audit-logs` from being rewritten as `logs`.
+- Canonical target identity is now validated independently of legacy-ID proof: an unproven or historical directory name that is not lowercase kebab-case cannot become a new Spec path until the caller supplies an explicit stable topic slug.
+- Legacy-ID proof now normalizes catalog-relative paths after Markdown unwrapping and searches reachable catalog history, so the lifecycle-only catalog transition does not make existing ID-prefixed directories impossible to hygienically rename.
+
+## Legacy Change Log
+
+- No legacy change-log section was recorded for this spec.
+- Legacy compatibility identity: `#weppn`.
+
+## References
+
+- `./SPEC.md`
+- `./IMPLEMENTATION.md`

package/toolkit/skills/docs/vendor/dot-codex/specs/weppn-topic-spec-model/IMPLEMENTATION.md

@@ -0,0 +1,66 @@
+# 主题级 Spec 模型与历史 specs hygiene 实现状态
+
+> 当前有效规范仍以 `./SPEC.md` 为准；这里记录实现覆盖、交付进度与 rollout 相关事实，避免这些细节散落到 PR / Git 历史里。
+
+## Current Status
+
+- Implementation: 已实现
+- Lifecycle: active
+- Catalog note: 让 `docs/specs/**` 回归主题级规格，并补齐历史 hygiene / workflow / migrate 约束
+- Current rollout focus: companion-doc 模型已落地，现有 `.codex` specs 已补 `IMPLEMENTATION.md` / `HISTORY.md`，README 也已恢复轻量 implementation 摘要。
+
+## Historical rollout snapshot
+
+- No pre-companion rollout snapshot was recorded for this spec.
+- Current lifecycle: active
+- Catalog note: 让 `docs/specs/**` 回归主题级规格，并补齐历史 hygiene / workflow / migrate 约束
+
+## Delivery Milestones
+
+- [x] topic-level `SPEC.md` 与 `spec_disposition` 模型已建立
+- [x] `spec-sync` / workflow 契约已切到 companion-doc 模型
+- [x] 现有 `.codex` specs 已完成一轮 companion-doc backfill
+- [x] `docs-migrate MODE=specs-hygiene` 已改为“迁移细节”，不再默认删除细节
+- [x] legacy `PLAN.md` sync 与 `MODE=specs` bootstrap 已补齐 companion-doc 边界，不会让 legacy-plan-only 仓库被误判 drift
+- [x] 历史 `SPEC.md`-only canonical spec 目录保留读取兼容；只有已出现 companion docs 的目录才要求三件套完整
+- [x] 新建 Spec 与 assets/metadata 已切至 slug-only；historical ID-prefixed Spec 继续只读兼容，hygiene 迁移需要显式 rename/delete confirmation
+- [x] Spec drift 的 legacy ID 去前缀已与 docs-migrate 的 proof 规则对齐：必须有历史 catalog 的精确 `ID + source path` 或同一 diff Git rename
+- [x] docs-migrate guard 仅接受 lowercase kebab-case canonical topics；无法直接作为 slug 的 legacy 目录必须由调用方通过 `--topic` 明确命名
+- [x] legacy ID proof 支持从可达 catalog 历史读取相对路径、inline-code 与 Markdown-link 形式，避免已切换 lifecycle catalog 后失去历史目录规范化能力
+
+## Rollout Artifacts
+
+### Docs touched
+
+- `AGENTS.md`
+- `docs/README.md`
+- `docs/specs/README.md`
+- `skills/docs-plan/**`
+- `skills/spec-sync/**`
+- `skills/project-engineering-workflow/SKILL.md`
+- `skills/docs-migrate/**`
+- `docs/specs/topic-spec-model/**`
+
+### Plan assets
+
+None
+
+### Asset promotion
+
+None
+
+## Related Changes
+
+- `b39a914` docs(specs): adopt topic-level spec model
+- `b21cbfe` feat(skills): add solution sync and spec sync
+- `6ecce09` fix(spec-sync): honor legacy spec refs
+- `1ce2aff` fix(review): block incomplete associated specs
+
+## Remaining Gaps
+
+- 个别历史 spec 的 `HISTORY.md` 仍可能只先保留 commit trace；后续若主题演进继续复杂化，应补更细的人工摘要。
+
+## References
+
+- `./SPEC.md`
+- `./HISTORY.md`

package/toolkit/skills/docs/vendor/dot-codex/specs/weppn-topic-spec-model/SPEC.md

@@ -0,0 +1,178 @@
+# 主题级 Spec 模型与历史 specs hygiene
+
+> 当前有效规范以本文为准；实现覆盖与当前状态见 `./IMPLEMENTATION.md`，关键演进原因见 `./HISTORY.md`。
+
+## 背景 / 问题陈述
+
+- 当前 `docs/specs/**` 虽然承担工程流程真相源，但长期被写成“一任务一张工单”：包含 `Status`、`Milestones`、`Docs to Update`、`Change log` 等任务推进字段。
+- 这种写法会让 spec 既不像真正的规格，也承担不了长期主题契约的职责；人和 Agent 都会把它误读成任务单。
+- 前一轮已经补上了 `project_doc_disposition`、`$project-doc-promote` 与 `MODE=project-docs`，但那只解决了“稳定项目真相如何 promotion 给人类文档”，没有解决 `docs/specs/**` 本体失真。
+
+## 目标 / 非目标
+
+### Goals
+
+- 把 `docs/specs/**` 重定义为 topic-level specification：一个 spec 描述一个稳定主题、能力、行为、接口或约束。
+- 新建 spec 使用唯一 slug-only 目录 `docs/specs/<topic>/`；历史 ID-prefixed 目录仅保留读取兼容，规范化由明确确认的 hygiene 迁移完成。
+- 让每个 spec 目录同时保留三层信息：`SPEC.md`（当前规范）、`IMPLEMENTATION.md`（实现状态/覆盖进度）、`HISTORY.md`（关键演进原因）。
+- 把共享流程的 spec gate 改成 `spec_disposition=link|update|create|supersede|not_needed`，不再默认“一任务一 spec”。
+- 保留 spec 中的验收 checklist、`Related PRs` 与 `Visual Evidence`，并把实现状态 / 变更原因从单次任务 prose 提炼进 companion docs，而不是把这些信息删掉。
+- 扩展 `$docs-migrate` 的 `MODE=specs-hygiene`，支持历史 `docs/specs/**` 的批量回填与收敛。
+- 对当前 `.codex` 仓的现有 `docs/specs/**` 做一轮 hygiene：把 `Status` / `Milestones` / `Change log` 等有价值细节迁到 companion docs，统一 README lifecycle + 轻量 implementation 摘要，并保留真正的主题规范内容。
+
+### Non-goals
+
+- 不把任务进度重新塞回 `SPEC.md` 正文的其它别名字段。
+- 不把 `docs/solutions/**` 或 human-facing project docs 重新并入 `docs/specs/**`。
+- 不在这轮实现里为所有外部项目仓库做批量自动改造。
+
+## 范围（Scope）
+
+### In scope
+
+- `AGENTS.md`、`docs/README.md`、`docs/specs/README.md`、`docs/solutions/README.md` 对 spec 模型的全局描述。
+- `skills/docs-plan/**`、`skills/project-engineering-workflow/SKILL.md`、`skills/spec-sync/SKILL.md`（以及 `plan-sync` compatibility alias）的 topic-spec 契约。
+- `skills/pr-create/SKILL.md`、`skills/pr-merge/SKILL.md`、`skills/github-pr-body-images/SKILL.md` 的 PR 元数据约定。
+- `skills/docs-migrate/**` 的 `specs-hygiene` 轨与 `auto` 分流。
+- `skills/find-next-plan/**` 对 lifecycle-based spec 目录的兼容行为。
+- 当前 `.codex` 仓现有 `docs/specs/**` 的 companion-doc 回填、README implementation 摘要回填与 README lifecycle 收敛。
+
+### Out of scope
+
+- 自动把 topic spec 再拆成强制统一的产品文档树。
+- 为主题级 spec 引入新的 id 体系或额外索引文件。
+- 在没有明确证据时自动合并多个历史 spec 为一个新 spec。
+
+## 需求（Requirements）
+
+### MUST
+
+- `docs/specs/**` 必须被定义成 topic-level specification，而不是 work-item contract / task ticket。
+- 每个 spec 目录默认必须由 `SPEC.md + IMPLEMENTATION.md + HISTORY.md` 共同承载规范、实现状态与关键演进原因。
+- 共享流程在进入实现前必须判定 `spec_disposition=link|update|create|supersede|not_needed`。
+- 当 `spec_disposition != not_needed` 时，关联 spec 必须至少覆盖背景、目标、范围、需求/契约与验收。
+- `docs/specs/README.md` 必须表达 `active|superseded|archived` 生命周期，并通过 `Successor` 列指向后继路径；README 不能重新变回完整任务看板。
+- 新建 canonical Spec、solution `related_specs`、视觉证据和 PR metadata 必须使用 `docs/specs/<topic>/...`；ID 仅可出现在明确标注的 legacy compatibility 路径或 fixture。
+- 默认 spec 模板必须把 `Status`、`Milestones`、`Docs to Update`、`Plan assets`、`Asset promotion` 等 rollout / progress 信息放进 `IMPLEMENTATION.md`，把 `Change log` / 关键演进原因放进 `HISTORY.md`，而不是简单删除。
+- spec 可以保留 `Acceptance checklist`、`Related PRs`、`Visual Evidence`，但必须明确这些内容服务于规格定义，而不是任务追踪。
+- `pr-create` / `pr-merge` / `github-pr-body-images` 必须改用 `Spec Disposition` 作为当前 PR 与 spec 的关联语义；旧 `Spec Status` / `Plan Status` 仅保留读取兼容。
+- canonical spec drift skill 名称必须是 `$spec-sync`；`$plan-sync` 只保留兼容别名，不再作为新写法输出。
+- `$spec-sync` 在读取 legacy `docs/plan/**/PLAN.md` 引用时，若仓库已经 canonicalize 到 `docs/specs/**`，必须优先按显式路径、同一 diff Git rename，或由历史 catalog 中精确 `ID + source path` 证明的唯一 slug 映射到 owning canonical `SPEC.md`；只有当必须选择 canonical target 却无法证明或无法唯一映射时才阻断并要求 canonical 路径，不能要求历史 `PLAN.md` 也同步变化。仍是独立真相源的 readable legacy `PLAN.md` 保持合法同步目标。
+- 对保留的 legacy ID-prefixed Spec 只读兼容目录，若 lifecycle catalog 已切换为 slug-only，可在 owning canonical ``HISTORY.md`` 使用精确的 ``- Legacy compatibility identity: \``#id\`.` 标记作为映射证明；任意普通文本出现 ID 都不构成证明。
+- 若仓库尚未 canonicalize、legacy `docs/plan/**/PLAN.md` 仍是唯一 spec 真相源，`$spec-sync` 必须继续把该 `PLAN.md` 视为合法同步目标；不能因为 companion-doc 模型而把 legacy plan-only 仓库全部误判成 drift。
+- 对已经 canonicalize 到 `docs/specs/**`、但历史上只有 `SPEC.md` 的旧 spec 目录，`$spec-sync` 必须继续把该 `SPEC.md` 视为合法同步目标；只有当目录里已经出现任一 companion doc 时，才要求 `IMPLEMENTATION.md` / `HISTORY.md` 也保持完整。
+- 对 `spec_disposition=not_needed` / spec-free 运行，只有在当前分支未触碰任何 `docs/specs/**` 或 legacy `docs/plan/**` 时才允许返回 `none(spec-free)`；若 spec 文件也发生变化，必须视为阻断而不是自动放行。
+- `$docs-migrate` 必须新增 `MODE=specs-hygiene`，用于历史 `docs/specs/**` 的批量整理。
+- `$docs-migrate MODE=specs` 在 legacy-plan-only 场景下建立 slug-only canonical `docs/specs/<topic>/` 时，必须一次补齐 `SPEC.md`、`IMPLEMENTATION.md`、`HISTORY.md`，避免刚完成 canonicalization 就立刻被 companion-doc drift gate 阻断。
+- 当前 `.codex` 仓现有 spec 必须完成至少一轮 companion-doc backfill，不得只更新未来模板而不处理现有 spec。
+- 当仓库已经建立 lifecycle-based `docs/specs/**`，但仍保留 `docs/plan/**` 作为过渡期 task queue 时，`find-next-plan` 必须继续能从 legacy queue 里选到下一个任务，而不是被 lifecycle index 提前短路。
+- 对 spec-free / `spec_disposition=not_needed` 的任务，`goto-impl-branch` 不得为了切 topic branch 而伪造新的 `docs/specs/**`。
+
+### SHOULD
+
+- lifecycle-based `docs/specs/README.md` 应继续保留稳定创建顺序，便于回溯主题演进。
+- `docs/specs/README.md` 的 `Implementation` 列应保持短小、稳定、可快速扫描；详细状态不要重新塞回 README。
+- 若历史 spec 仍残留 legacy `## Visual Evidence (PR)` heading，应统一回到单一 `## Visual Evidence`。
+- `find-next-plan` 在 topic-spec 仓库里应明确报告“spec 目录不再编码任务队列”，而不是误选一个 `active` spec 当任务单。
+
+## 功能与行为规格（Functional/Behavior Spec）
+
+### Core flows
+
+- 共享工程流程的 Step 2 先解决“当前任务与哪个 spec 的关系是什么”，而不是直接为任务建卡。
+- `spec_disposition=link`：当前任务实现或引用既有 topic spec，不改变 spec contract。
+- `spec_disposition=update`：当前任务改变了既有 topic spec 的 contract，需要先更新 spec 再实现。
+- `spec_disposition=create`：当前任务引入新的稳定主题/契约，需要创建新的 topic spec。
+- `spec_disposition=supersede`：当前任务用新 spec 取代旧 spec；README lifecycle 必须同步标记后继关系。
+- `spec_disposition=not_needed`：当前任务不改变任何稳定 contract，也不需要依赖 spec 作为主规范输入。
+- `MODE=specs-hygiene` 用于批量盘点既有 `docs/specs/**`，把旧 task-ticket section 中仍有价值的实现状态 / rollout 事实 / 变更原因迁入 companion docs，统一 lifecycle，并在需要时协同 `MODE=project-docs` 做人类文档 promotion。
+
+### Edge cases / errors
+
+- 若历史仓库仍依赖 legacy `docs/plan/**`，先通过 `MODE=specs` 建立 canonical `docs/specs/**`，再进入 `specs-hygiene`；不得跳过 canonical root。
+- 若某个历史 spec 只剩任务过程而缺少稳定契约，应把它标记为 `archived` 或 `superseded` 并设置 `Successor`，而不是继续留作 `active`。
+- 若同一轮既涉及新建 spec 又涉及旧 spec lifecycle 变更，README lifecycle 必须与实际 spec 路径同时更新，避免出现 dangling 引用。
+
+## 接口契约（Interfaces & Contracts）
+
+### 接口清单（Inventory）
+
+| 接口（Name） | 类型（Kind） | 范围（Scope） | 变更（Change） | 契约文档（Contract Doc） | 负责人（Owner） | 使用方（Consumers） | 备注（Notes） |
+| - | - | - | - | - | - | - | - |
+| spec_disposition | workflow field | internal | New | None | codex-global | shared workflow / PR flows | topic-spec gate |
+| spec-sync | skill | internal | New | None | codex-global | shared workflow / PR flows | canonical spec drift / sync 入口 |
+| docs/specs lifecycle index | docs catalog | internal | Modify | `docs/specs/README.md` | codex-global | specs authors / agents | `active | superseded | archived` |
+| docs-migrate MODE=specs-hygiene | skill mode | internal | New | `skills/docs-migrate/references/specs-hygiene-mode.md` | codex-global | docs migration | 历史 specs 批量回填 |
+| Spec Disposition PR metadata | PR body contract | internal | Modify | None | codex-global | `pr-create` / `pr-merge` / `github-pr-body-images` | 替代 `Spec Status` |
+
+## 验收标准（Acceptance Criteria）
+
+- Given 一个 topic-spec 仓库，When 创建新 spec，Then 它会同时生成 `SPEC.md`、`IMPLEMENTATION.md`、`HISTORY.md`，并在 `docs/specs/README.md` 追加 `Lifecycle=active` + 轻量 `Implementation` 摘要。
+- Given 一个任务只是在实现既有主题规范，When 进入共享 Step 2，Then 它会记录 `spec_disposition=link`，而不是默认新建任务 spec。
+- Given 一个任务改变了稳定 contract，When 进入共享 Step 2，Then 必须先完成 `update` 或 `create` 对应的 spec 更新，之后才能进入实现。
+- Given 一个历史 `SPEC.md` 仍包含 `Status`、`Milestones`、`Docs to Update` 等任务单 section，When 运行本轮 hygiene，Then 这些 section 的细节会被迁入 `IMPLEMENTATION.md` / `HISTORY.md`，而不是直接丢弃。
+- Given 一个 Agent 只打开 spec 目录，When 它需要回答“现在做到哪了”或“为什么会变成这样”，Then 它可以直接从 `IMPLEMENTATION.md` / `HISTORY.md` 获取答案，而不必跨 PR / commit 历史搜索。
+- Given 一个历史 spec 仍保留 `## Visual Evidence (PR)`，When 迁到新模型，Then 最终只保留单一 `## Visual Evidence` heading。
+- Given 一个 lifecycle-based `docs/specs/README.md`，When 调用 `find-next-plan`，Then 它会返回“该目录不再编码任务队列”的明确结果，而不是误把 `active` spec 当成待办工单。
+- Given 一个仓库已经有 lifecycle-based `docs/specs/README.md`，但 `docs/plan/README.md` 仍保留 legacy task queue，When 调用 `find-next-plan`，Then 它会回退到 legacy queue 继续选任务，并优先解析到 canonical `docs/specs/**/SPEC.md`。
+- Given 一个合并前 PR，When `pr-create` / `pr-merge` 处理 spec metadata，Then 新正文使用 `Spec Disposition`，旧 `Spec Status` 只作为读取兼容。
+- Given 旧会话或旧文案仍显式触发 `$plan-sync`，When 进入 spec drift gate，Then 它会转到 `$spec-sync` 的同一 tri-state 契约，而不是维护第二套规则。
+- Given 调用方仍传入 legacy `docs/plan/<legacy-id>-*/PLAN.md` 作为 spec 引用，When 仓库已经把该主题迁到唯一的 canonical `docs/specs/<topic>/SPEC.md`，Then drift checker 仍会把 canonical `SPEC.md` 的变更视为有效同步，而不是错误阻断。
+- Given 一个合法 slug 形如 `audit-logs`，When 没有历史 catalog 的精确 `ID + source path` 记录且没有同一 diff Git rename，Then drift checker 不得把它按五字符前缀重写成 `docs/specs/logs/SPEC.md`。
+- Given 一个仓库仍只有 legacy `docs/plan/<legacy-id>-*/PLAN.md`，When 当前实现与该 `PLAN.md` 一起更新、但尚未建立 canonical `docs/specs/**`，Then drift checker 仍会把该 `PLAN.md` 视为有效同步目标，而不是因为 companion docs 缺失直接阻断。
+- Given 一个仓库仍保留历史 ID-prefixed `docs/specs/<legacy-id>-*/SPEC.md`，且该目录仍是 `SPEC.md`-only 形态，When 当前实现与该 `SPEC.md` 一起更新、且尚未做 companion-doc backfill，Then drift checker 仍会把该 `SPEC.md` 视为有效同步目标，而不是要求先补齐 `IMPLEMENTATION.md` / `HISTORY.md`。
+- Given 一个 legacy-plan-only 仓库通过 `$docs-migrate MODE=specs` 建立 canonical `docs/specs/**`，When canonicalization 完成后进入下一轮 Step 2 / drift gate，Then 新建的 canonical spec 目录已经包含 `SPEC.md`、`IMPLEMENTATION.md`、`HISTORY.md`，不会被“associated spec files are missing”立即打回。
+- Given 旧计划目录名不是 lowercase kebab-case，When `$docs-migrate MODE=specs` 或 `MODE=specs-hygiene` 生成迁移映射，Then guard 必须阻断并要求显式的稳定 `--topic`，不得直接生成非 slug canonical Spec 路径。
+- Given 一个历史 catalog 使用相对 catalog 目录的 inline-code 或 Markdown-link `Spec` 路径，When `$docs-migrate MODE=specs-hygiene` 证明 legacy ID 前缀，Then guard 必须将该路径规范化并可从可达 Git 历史读取精确 `ID + path` 证据。
+- Given 一个任务被判定为 `spec_disposition=not_needed`，When 当前分支同时改动了实现文件和 `docs/specs/**`，Then drift checker 必须返回阻断并指出 spec-free 与 spec 改动冲突。
+- Given 一个仓库既没有 `docs/specs/**` 也没有 `docs/plan/**`，When `goto-impl-branch` 用于明确为 `spec_disposition=not_needed` 的实现任务，Then 它会直接创建 topic branch 而不是伪造新的 spec 根。
+
+## 验收清单（Acceptance checklist）
+
+- [x] `docs/specs/**` 的全局定义已改为 topic-level spec。
+- [x] 共享流程已改用 `spec_disposition`，不再依赖 spec 完成状态。
+- [x] spec 模板与 README index 已改成“lifecycle + 轻量 implementation”，并默认生成 companion docs。
+- [x] `docs-migrate` 已新增 `MODE=specs-hygiene`，并补充最小样例。
+- [x] `.codex` 仓现有 `docs/specs/**` 已完成至少一轮 companion-doc backfill。
+- [x] PR body / merge gate / image rewrite 契约已切到 `Spec Disposition`。
+
+## 非功能性验收 / 质量门槛（Quality Gates）
+
+### Testing
+
+- `git diff --check`
+- `python3 /Users/ivan/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/docs-plan skills/docs-migrate skills/find-next-plan skills/pr-create skills/pr-merge skills/github-pr-body-images skills/project-engineering-workflow skills/spec-sync skills/plan-sync`
+- `python3 -m unittest skills/find-next-plan/scripts/test_find_next_plan.py`
+- `python3 -m unittest skills/spec-sync/scripts/test_spec_drift_check.py`
+- `rg -n "IMPLEMENTATION.md|HISTORY.md|Implementation \\||MODE=specs-hygiene|spec_disposition" AGENTS.md docs skills`
+
+### Quality checks
+
+- active spec 的 `SPEC.md` 不再包含任务单 section，但同目录必须存在可读的 `IMPLEMENTATION.md` / `HISTORY.md`。
+- lifecycle index 与 spec 目录实际路径一致。
+- legacy 兼容只保留读取路径，不再作为新写法输出。
+
+## Visual Evidence
+
+None
+
+## Related PRs
+
+- None（local-merge path）
+
+## 风险 / 开放问题 / 假设（Risks, Open Questions, Assumptions）
+
+- 风险：历史 spec 的 companion-doc 回填仍可能先以“提取旧字段 + commit trace”起步，个别主题后续还需要补更细的人工梳理。
+- 风险：旧 PR/脚本仍可能携带 `Spec Status` 元数据；因此这轮只把它降级为读取兼容，而不是一次性抹掉所有历史痕迹。
+- 假设：当前 `.codex` 仓现有 spec 基本都还对应有效主题，不需要本轮大规模做 `archived` / `superseded` 生命周期分流。
+
+## 参考（References）
+
+- `docs/specs/agent-memory-project-docs/SPEC.md`
+- `docs/specs/global-solutions-docs-migrate/SPEC.md`
+- `skills/docs-plan/SKILL.md`
+- `skills/project-engineering-workflow/SKILL.md`
+- `skills/docs-migrate/SKILL.md`
+- `skills/spec-sync/SKILL.md`
+- `skills/pr-create/SKILL.md`
+- `skills/pr-merge/SKILL.md`

package/toolkit/skills/docs/vendor/dot-codex/specs/workflow-agent-model-selection/HISTORY.md

@@ -0,0 +1,29 @@
+# History
+
+- The flow-scoped policy now requires a persisted Wave Thread task class with every locked profile, keeps pre-policy v2 state read-only until explicit migration, rejects stale parent bases and duplicate dispatch claims, and preserves historical profile events across an explicit owner replacement while rejecting new unbound owner-override events.
+
+- The shared project-engineering direct-review contract now defines the exact six-profile autonomous allowlist, lane-local selection, complete-pair owner overrides scoped to an invocation or named lane, and runtime-owned rejection without silent fallback.
+
+- Flow-specific profile policies were agreed for Code Review and Wave Thread: each flow has an independent allowlist, Code Review lanes choose independently, Wave Thread tasks lock one profile, and a direct complete owner-specified pair may bypass the corresponding allowlist while runtime support remains authoritative.
+
+- Direct review moved from receipt-enforced profile ladders to lane-level guidance and direct structured evidence.
+- Formal review launch is now readiness-gated and runtime-created per current Candidate; the plan's profile is advisory and terminal reviewers are never reused.
+- Readiness now covers acceptance, scope, required contracts, and validation across the local-merge consumer; direct plans must align strategy/profile guidance with their selected tier, and structured lane evidence must bind exact scope and a non-empty actual profile without unknown fields.
+- The direct-plan checker now verifies that Tier 3/4 recommendations name a concrete supported model/effort pair and that runtime strategy text includes an explicit formal-lane launch/create action.
+- Structured replay now rejects unknown disposition values before any repair or report-only classification.
+- Structured replay now requires explicit readiness and a demonstrated finding before initial formal launch or In-Scope repair.
+- Structured replay now rejects unsupported actual profiles and unknown dispositions embedded in terminal findings.
+- Structured replay now distinguishes demonstrated findings from observations before authorizing an In-Scope repair.
+- Structured replay rejects missing or unknown finding kinds before terminal clear state.
+- Valid terminal clear evidence may report a blocking finding; unavailable or malformed review output alone is `unchecked`.
+
+- 当前合同将 transcript/profile attestation 降为历史兼容信息。显式启动 profile 仍做低成本输入校验，PTC 与 direct 传输等价；direct-flow review proof 改由 helper-owned structured `ReviewReceipt` 决定，advisory review 不能替代该 proof。
+
+- Workflow agent models were previously treated as exact stage and lane requirements. That made review and delegated work unavailable when a recommended override could not be launched.
+- The contract now separates preferred planning profiles from resolved runtime profiles and permits one bounded explicit replacement fallback before startup.
+- Native v3 review evidence now records the tool-observed resolved model and reasoning effort, and rejects missing or mismatched task-envelope values before accepting a lane receipt.
+- The v3 envelope was still caller-controlled and could infer a fallback from a preferred/resolved mismatch. Schema v4 replaces that inference with a common native launch hook, one-time pre-start permit, and transcript-derived runtime attestation.
+
+## Wave Thread 移除
+
+Wave Thread 专属三档 Profile 与派发回放已经删除。直接 Review 和现行多任务流程的 Profile 规则继续由各自合同负责。

package/toolkit/skills/docs/vendor/dot-codex/specs/workflow-agent-model-selection/IMPLEMENTATION.md

@@ -0,0 +1,42 @@
+# Workflow Agent Model Selection Implementation
+
+- The shared project-engineering direct-review contract now exposes the exact six-profile autonomous allowlist, independent lane-local selection, complete invocation-wide or named-lane owner overrides, and runtime-owned rejection without silent fallback.
+
+- `bin/modules/flow/review-profile-policy.ts` owns direct-review complexity mapping and exact owner-pair forwarding without importing the runtime capability table.
+- The direct-review selector requires lane-specific reasons for special or high-risk profiles and resolves invocation-wide versus named-lane overrides across the complete lane set.
+- `bin/modules/flow/profile-hook.ts` validates explicit supported launch profiles and preserves native request identity aliases.
+- The launch hook keeps autonomous selection to the four allowed model families and records only a session-scoped explicit model directive from the first non-empty `UserPromptSubmit` line.
+- Direct review results record the selected profile and any high-effort reason.
+- Historical coordinator task dispatch remains a separate compatibility contract: its task lock is forwarded verbatim, native-create request identity is bound independently of runtime capability support, and alias conflicts or deferred recovery context corruption fail closed.
+- Runtime formal review first checks Review Readiness; plans carry only a profile recommendation, while each changed Candidate gets a fresh reviewer.
+- Shared review routing does not use the flow receipt adapter to authorize profiles, concurrency, or evidence acceptance.
+- This policy does not restore, refactor, or reclassify the retired review orchestration ladder; that retired boundary and its historical compatibility facts remain unchanged.
+- Review findings use the shared In-Scope, Design-Returning, and Observation dispositions; only the first enters a bounded Repair Batch.
+- Profile/recovery tests bind every continuation anchor field, including lane, scope, current locked flow, and cumulative Repair Batch count, before replay.
+- Active Spec regression tests cover the shared disposition, `unchecked` evidence, and complete anchor contract in the model-selection topic.
+- The same isolated-copy suite rejects retired review dispositions in this active Spec.
+- ReviewReplay rejects missing dispositions and incomplete owner authorization anchors before mutating review state.
+- The instruction architecture checker guards active review Specs, anchor boundaries, and consumer vocabulary against retired contract terms.
+- The architecture checker also guards the readiness contract and ADR 0017 binding in isolated copies.
+- Direct plan validation checks tier-consistent strategy/profile guidance, and ReviewReplay validates exact scope, non-empty actual profile, and unknown fields before terminal state.
+- Direct plan validation requires Tier 3/4 strategy text to state a runtime formal-lane launch/create action and requires profile guidance to name a concrete supported model/effort pair; runtime reviewer identity stays fresh and dynamic.
+- ReviewReplay rejects unknown finding dispositions before repair classification, preserving the fail-closed boundary for malformed results.
+- ReviewReplay requires explicit readiness and demonstrated findings before initial launch or In-Scope repair.
+- ReviewReplay validates the actual model/effort pair against the runtime's supported profile table and rejects unknown embedded finding dispositions.
+- ReviewReplay requires the finding kind to be demonstrated before an In-Scope Repair Batch may mutate the candidate.
+- Terminal structured findings are restricted to the explicit demonstrated or observation kinds.
+
+## Verification
+
+- `bun test bin/tests/flow bin/tests/waves`
+- `bun run bin/wave_thread_flow.ts --compatibility-smoke`
+- `bun run bin/compatibility_cli.ts wave_thread_replay_check`
+- `bun run bin/compatibility_cli.ts wave_thread_roundtrip`
+- `bun run bin/compatibility_cli.ts spec_contract_check --path docs/specs/wave-thread-flow/SPEC.md --repo-root .`
+- `bun run bin/compatibility_cli.ts skills_index_check`
+- `bun check`
+- `git diff --check`
+
+## Wave Thread 移除
+
+Wave Thread 专属三档 Profile 与派发回放已经删除。直接 Review 和现行多任务流程的 Profile 规则继续由各自合同负责。