flower-trellis 0.6.0-beta.0 → 0.6.0-beta.2

package/enhancements/0.6/.agents/skills/trellis-check-all/SKILL.md

@@ -18,7 +18,7 @@ description: "统一 Check-All 入口：确认范围与运行上下文，按 req
 4. 全程收集普通问题到 `CHK-*`，收集低风险文档漂移到 `DOC-*`。
 5. 在最终报告前处理允许自动修复的文档漂移，并把修复内容展示在报告里。
 6. 根据 interactive / validated auto-loop 边界输出下一步或完成 runner `record + next`。
-7. untracked 上下文在最终 diff 稳定后调用 `untracked_flow.py record-check`；只有严格通过且 disposition 确认继续时才 `advance --stage spec`。
+7. untracked helper 只保存流程游标：findings 或新编辑设回 `implement`；只有严格通过且 disposition 确认继续时才 `advance --stage spec`。
 
 ---
 

package/enhancements/0.6/.agents/skills/trellis-check-all/references/depth-routing.md

@@ -37,9 +37,9 @@ git log --oneline -10
 
 没有当前 task 时运行 `python3 ./.trellis/scripts/untracked_flow.py status --verbose`：
 
-- `hit`：读取 work id、summary、stage、scope、baseline/current fingerprint 和已有验证证据；三件套实现维度标记 `N/A`，其余维度仍对实际 diff 与相关 spec 负责。
+- `hit`：读取 work id、summary 和 stage；三件套实现维度标记 `N/A`，其余维度仍对实际 diff、相关 spec 和本轮可验证证据负责。
 - `miss`：仅当用户明确要求检查一个无状态的已知 diff 时继续，并把工作上下文缺失列为风险；否则停止并回到 Request Triage。
-- `error` 或 workspace drift：按阻塞报告，禁止用聊天摘要恢复或覆盖状态。
+- `error`：按损坏状态阻塞报告，禁止用聊天摘要恢复或覆盖游标。
 
 untracked 检查必须处于 `stage=check`，且只读取个人 check 偏好，不创建 task-scoped route decision。
 

package/enhancements/0.6/.agents/skills/trellis-check-all/references/full-profile.md

@@ -6,7 +6,7 @@ Full 是完整验收映射和全影响面审查。只有 `check_profile.effectiv
 
 ## Step 1：对照规划三件套检查实现
 
-untracked 上下文没有 task artifacts，本 Step 标记 `N/A`；不得把 summary、scope 或聊天记录当成 PRD。仍须完整执行 Step 2 与 Step 3，并对实际 diff、相关 spec、状态机证据和多仓分发边界负责。
+untracked 上下文没有 task artifacts，本 Step 标记 `N/A`；不得把 summary、stage 或聊天记录当成 PRD。仍须完整执行 Step 2 与 Step 3，并对实际 diff、相关 spec、当前验证证据和多仓分发边界负责。
 
 ### 1.1 验收依据
 

package/enhancements/0.6/.agents/skills/trellis-check-all/references/reporting-and-disposition.md

@@ -125,7 +125,7 @@ interactive 模式完成所有可继续检查和允许的 `DOC-*` 自动修复
 
 检查通过后的动作由下方 `Interactive Post-Check Stop Gate` 判断：普通交互停止等待，符合 direct Git 严格通过条件时同轮进入 Phase 3.3 `trellis-update-spec`，再到 Phase 3.4 `trellis-push`。仍有 `CHK-*` 时停留在修复/重检循环。
 
-untracked 在最终报告前调用 `untracked_flow.py record-check`：严格通过记录 `pass`，有问题记录 `findings`，部分验证记录 `partial`，真正阻塞记录 `blocked`。普通严格通过但尚未继续时保持 `stage=check`；只有 direct Git 同轮继续或用户后续明确继续时才 `advance --stage spec`。任何报告后的新编辑先回到 `prepare-edit`，旧检查证据随 fingerprint 失效。
+untracked helper 不记录 Check-All 证据。普通严格通过但尚未继续时保持 `stage=check`；只有 direct Git 同轮继续或用户后续明确继续时才 `advance --stage spec`。有 findings、部分验证、阻塞或报告后的新编辑时，先 `advance --stage implement` 再返回实现。
 
 ---
 

package/enhancements/0.6/.agents/skills/trellis-push/SKILL.md

@@ -17,7 +17,7 @@ description: "按确认的精确文件范围提交普通变更或完成已就绪
 - 不发起、终止或解决分支合并；只允许普通模式完成已经开始、冲突已清零且索引完全可归属的 merge commit。
 - 不处理上线核对、任务归档、会话日志或自动任务队列状态。
 - 不使用 `git add .`、`git add -A`，不要求工作区整体干净，也不提交计划外文件。
-- untracked 上下文只接受 `stage=push` 且 Check-All / Update-Spec 证据与当前 workspace fingerprint 一致；不生成任务进度提交。
+- untracked 上下文只接受 `stage=push`；该状态只负责路由，不替代本 skill 的正式计划、确认和 Git 安全检查，也不生成任务进度提交。
 
 ## 模式
 
@@ -40,7 +40,7 @@ description: "按确认的精确文件范围提交普通变更或完成已就绪
 
 auto-loop 内部 `commit-only` 已由 runner 的 `run_check_all -> run_spec_update -> commit_only` 状态机和预授权保证顺序，因此不重复记录或判断本交互证据。
 
-没有活动 task 时运行 `python3 ./.trellis/scripts/untracked_flow.py status --verbose`。命中 untracked 后，以 helper 返回的 stage、scope、baseline、current fingerprint、Check-All 和 Update-Spec 证据填写上述完成链；必须为 `stage=push` 且证据仍有效。`miss` 才按既有“无活动任务”普通 Git 路径处理；`error` 或 workspace drift 停止，不从摘要猜测。
+没有活动 task 时运行 `python3 ./.trellis/scripts/untracked_flow.py status --verbose`。命中 untracked 后，记录 work id、summary 和 stage，并要求 `stage=push`；完成链证据仍从当前 Check-All / Update-Spec 结果与实际 diff 获取。`miss` 才按既有“无活动任务”普通 Git 路径处理；损坏状态停止，不从摘要猜测。游标命中不表示 Push 已计划、已确认或已执行。
 
 ## Step 1：发现仓库与任务
 
@@ -67,7 +67,7 @@ python3 ./.trellis/scripts/task_progress.py status --json || true
 git status --short --untracked-files=all -- <task-dir>
 ```
 
-不得把默认 `git status --short` 可能返回的 `?? <task-dir>/` 折叠目录当成 exact file、展示条目或 pathspec。无活动 task 时仍可提交相关代码，但不生成任务进度。untracked 命中时，所有业务 `planned` 文件必须能由当前 state scope 与实际 diff 归属，计划同时显示 work id；scope 外文件只能保留或作为归属风险。存在活动 task 时，结合 `brief.md`、`implement.md`、当前 diff 与本轮执行范围生成一行语义进度；同时识别当前任务目录中已存在且可归属的 dirty/untracked 产物，供 Step 5 生成任务记录 exact files。不得从旧进度推断 Git 动作。
+不得把默认 `git status --short` 可能返回的 `?? <task-dir>/` 折叠目录当成 exact file、展示条目或 pathspec。无活动 task 时仍可提交相关代码，但不生成任务进度。untracked 命中时，结合当前请求、work summary 和实际 diff 判断业务 `planned` 文件归属，计划同时显示 work id；无法明确归属的文件只能保留或作为风险。存在活动 task 时，结合 `brief.md`、`implement.md`、当前 diff 与本轮执行范围生成一行语义进度；同时识别当前任务目录中已存在且可归属的 dirty/untracked 产物，供 Step 5 生成任务记录 exact files。不得从旧进度推断 Git 动作。
 
 ## Step 2：预检与文件归属
 

package/enhancements/0.6/.agents/skills/trellis-route/SKILL.md

@@ -212,7 +212,7 @@ dispatch prompt 第一行始终遵守当前 task/untracked 契约。agent 文件
 
 ### Untracked Subagent Dispatch 契约
 
-untracked 的 implement/check subagent prompt 第一行固定为 `Untracked work: <work-id>`，并包含事项摘要、stage、scope、baseline 仓库摘要、current fingerprint、已有验证证据、相关 spec 路径和本轮明确职责。不得写 `Active task:`，不得要求 `prd.md`、`implement.jsonl` 或 `check.jsonl`。agent 必须直接执行，不得递归 dispatch implement/check agent。
+untracked 的 implement/check subagent prompt 第一行固定为 `Untracked work: <work-id>`，并包含事项摘要、当前 stage、实际 diff、相关 spec 路径、已有验证上下文和本轮明确职责。helper 只提供流程游标，不提供 scope、baseline、fingerprint 或 owner evidence。不得写 `Active task:`，不得要求 `prd.md`、`implement.jsonl` 或 `check.jsonl`。agent 必须直接执行，不得递归 dispatch implement/check agent。
 
 ### Subagent Check-All Dispatch 契约
 

package/enhancements/0.6/.claude/skills/trellis-check-all/SKILL.md

@@ -18,7 +18,7 @@ description: "统一 Check-All 入口：确认范围与运行上下文，按 req
 4. 全程收集普通问题到 `CHK-*`，收集低风险文档漂移到 `DOC-*`。
 5. 在最终报告前处理允许自动修复的文档漂移，并把修复内容展示在报告里。
 6. 根据 interactive / validated auto-loop 边界输出下一步或完成 runner `record + next`。
-7. untracked 上下文在最终 diff 稳定后调用 `untracked_flow.py record-check`；只有严格通过且 disposition 确认继续时才 `advance --stage spec`。
+7. untracked helper 只保存流程游标：findings 或新编辑设回 `implement`；只有严格通过且 disposition 确认继续时才 `advance --stage spec`。
 
 ---
 

package/enhancements/0.6/.claude/skills/trellis-check-all/references/depth-routing.md

@@ -37,9 +37,9 @@ git log --oneline -10
 
 没有当前 task 时运行 `python3 ./.trellis/scripts/untracked_flow.py status --verbose`：
 
-- `hit`：读取 work id、summary、stage、scope、baseline/current fingerprint 和已有验证证据；三件套实现维度标记 `N/A`，其余维度仍对实际 diff 与相关 spec 负责。
+- `hit`：读取 work id、summary 和 stage；三件套实现维度标记 `N/A`，其余维度仍对实际 diff、相关 spec 和本轮可验证证据负责。
 - `miss`：仅当用户明确要求检查一个无状态的已知 diff 时继续，并把工作上下文缺失列为风险；否则停止并回到 Request Triage。
-- `error` 或 workspace drift：按阻塞报告，禁止用聊天摘要恢复或覆盖状态。
+- `error`：按损坏状态阻塞报告，禁止用聊天摘要恢复或覆盖游标。
 
 untracked 检查必须处于 `stage=check`，且只读取个人 check 偏好，不创建 task-scoped route decision。
 

package/enhancements/0.6/.claude/skills/trellis-check-all/references/full-profile.md

@@ -6,7 +6,7 @@ Full 是完整验收映射和全影响面审查。只有 `check_profile.effectiv
 
 ## Step 1：对照规划三件套检查实现
 
-untracked 上下文没有 task artifacts，本 Step 标记 `N/A`；不得把 summary、scope 或聊天记录当成 PRD。仍须完整执行 Step 2 与 Step 3，并对实际 diff、相关 spec、状态机证据和多仓分发边界负责。
+untracked 上下文没有 task artifacts，本 Step 标记 `N/A`；不得把 summary、stage 或聊天记录当成 PRD。仍须完整执行 Step 2 与 Step 3，并对实际 diff、相关 spec、当前验证证据和多仓分发边界负责。
 
 ### 1.1 验收依据
 

package/enhancements/0.6/.claude/skills/trellis-check-all/references/reporting-and-disposition.md

@@ -125,7 +125,7 @@ interactive 模式完成所有可继续检查和允许的 `DOC-*` 自动修复
 
 检查通过后的动作由下方 `Interactive Post-Check Stop Gate` 判断：普通交互停止等待，符合 direct Git 严格通过条件时同轮进入 Phase 3.3 `trellis-update-spec`，再到 Phase 3.4 `trellis-push`。仍有 `CHK-*` 时停留在修复/重检循环。
 
-untracked 在最终报告前调用 `untracked_flow.py record-check`：严格通过记录 `pass`，有问题记录 `findings`，部分验证记录 `partial`，真正阻塞记录 `blocked`。普通严格通过但尚未继续时保持 `stage=check`；只有 direct Git 同轮继续或用户后续明确继续时才 `advance --stage spec`。任何报告后的新编辑先回到 `prepare-edit`，旧检查证据随 fingerprint 失效。
+untracked helper 不记录 Check-All 证据。普通严格通过但尚未继续时保持 `stage=check`；只有 direct Git 同轮继续或用户后续明确继续时才 `advance --stage spec`。有 findings、部分验证、阻塞或报告后的新编辑时，先 `advance --stage implement` 再返回实现。
 
 ---
 

package/enhancements/0.6/.claude/skills/trellis-push/SKILL.md

@@ -17,7 +17,7 @@ description: "按确认的精确文件范围提交普通变更或完成已就绪
 - 不发起、终止或解决分支合并；只允许普通模式完成已经开始、冲突已清零且索引完全可归属的 merge commit。
 - 不处理上线核对、任务归档、会话日志或自动任务队列状态。
 - 不使用 `git add .`、`git add -A`，不要求工作区整体干净，也不提交计划外文件。
-- untracked 上下文只接受 `stage=push` 且 Check-All / Update-Spec 证据与当前 workspace fingerprint 一致；不生成任务进度提交。
+- untracked 上下文只接受 `stage=push`；该状态只负责路由，不替代本 skill 的正式计划、确认和 Git 安全检查，也不生成任务进度提交。
 
 ## 模式
 
@@ -40,7 +40,7 @@ description: "按确认的精确文件范围提交普通变更或完成已就绪
 
 auto-loop 内部 `commit-only` 已由 runner 的 `run_check_all -> run_spec_update -> commit_only` 状态机和预授权保证顺序，因此不重复记录或判断本交互证据。
 
-没有活动 task 时运行 `python3 ./.trellis/scripts/untracked_flow.py status --verbose`。命中 untracked 后，以 helper 返回的 stage、scope、baseline、current fingerprint、Check-All 和 Update-Spec 证据填写上述完成链；必须为 `stage=push` 且证据仍有效。`miss` 才按既有“无活动任务”普通 Git 路径处理；`error` 或 workspace drift 停止，不从摘要猜测。
+没有活动 task 时运行 `python3 ./.trellis/scripts/untracked_flow.py status --verbose`。命中 untracked 后，记录 work id、summary 和 stage，并要求 `stage=push`；完成链证据仍从当前 Check-All / Update-Spec 结果与实际 diff 获取。`miss` 才按既有“无活动任务”普通 Git 路径处理；损坏状态停止，不从摘要猜测。游标命中不表示 Push 已计划、已确认或已执行。
 
 ## Step 1：发现仓库与任务
 
@@ -67,7 +67,7 @@ python3 ./.trellis/scripts/task_progress.py status --json || true
 git status --short --untracked-files=all -- <task-dir>
 ```
 
-不得把默认 `git status --short` 可能返回的 `?? <task-dir>/` 折叠目录当成 exact file、展示条目或 pathspec。无活动 task 时仍可提交相关代码，但不生成任务进度。untracked 命中时，所有业务 `planned` 文件必须能由当前 state scope 与实际 diff 归属，计划同时显示 work id；scope 外文件只能保留或作为归属风险。存在活动 task 时，结合 `brief.md`、`implement.md`、当前 diff 与本轮执行范围生成一行语义进度；同时识别当前任务目录中已存在且可归属的 dirty/untracked 产物，供 Step 5 生成任务记录 exact files。不得从旧进度推断 Git 动作。
+不得把默认 `git status --short` 可能返回的 `?? <task-dir>/` 折叠目录当成 exact file、展示条目或 pathspec。无活动 task 时仍可提交相关代码，但不生成任务进度。untracked 命中时，结合当前请求、work summary 和实际 diff 判断业务 `planned` 文件归属，计划同时显示 work id；无法明确归属的文件只能保留或作为风险。存在活动 task 时，结合 `brief.md`、`implement.md`、当前 diff 与本轮执行范围生成一行语义进度；同时识别当前任务目录中已存在且可归属的 dirty/untracked 产物，供 Step 5 生成任务记录 exact files。不得从旧进度推断 Git 动作。
 
 ## Step 2：预检与文件归属
 

package/enhancements/0.6/.claude/skills/trellis-route/SKILL.md

@@ -212,7 +212,7 @@ dispatch prompt 第一行始终遵守当前 task/untracked 契约。agent 文件
 
 ### Untracked Subagent Dispatch 契约
 
-untracked 的 implement/check subagent prompt 第一行固定为 `Untracked work: <work-id>`，并包含事项摘要、stage、scope、baseline 仓库摘要、current fingerprint、已有验证证据、相关 spec 路径和本轮明确职责。不得写 `Active task:`，不得要求 `prd.md`、`implement.jsonl` 或 `check.jsonl`。agent 必须直接执行，不得递归 dispatch implement/check agent。
+untracked 的 implement/check subagent prompt 第一行固定为 `Untracked work: <work-id>`，并包含事项摘要、当前 stage、实际 diff、相关 spec 路径、已有验证上下文和本轮明确职责。helper 只提供流程游标，不提供 scope、baseline、fingerprint 或 owner evidence。不得写 `Active task:`，不得要求 `prd.md`、`implement.jsonl` 或 `check.jsonl`。agent 必须直接执行，不得递归 dispatch implement/check agent。
 
 ### Subagent Check-All Dispatch 契约
 

package/enhancements/0.6/overrides/conflicts.json

@@ -375,8 +375,8 @@
           "| Request Intent Routing | `Request Triage` + `trellis-start` | `task_intent.py` |",
           "| Project Knowledge Discovery | `Request Triage` | `spec_router.py` |",
           "| Active Task Scope Guard | `Request Triage` | `task_intent.py` scope safety |",
-          "| Untracked Work Completion Chain | `workflow-state:untracked` + Phase 2/3 owners | `untracked_flow.py` |",
-          "| Untracked Task Adoption | `workflow-state:untracked` + `trellis-brainstorm` | `task_intent.py adopt` |",
+          "| Untracked Work Completion Chain | `workflow-state:untracked*` + Phase 2/3 owners | `untracked_flow.py` |",
+          "| Untracked Task Adoption | `workflow-state:untracked*` + `trellis-brainstorm` | `task_intent.py adopt` |",
           "| Routing Gate | Phase 2 + `trellis-route` | `route_state.py` |",
           "| Auto-Loop Return Gate | `trellis-check-all` + `trellis-auto-loop` | `auto_loop.py record/next` |",
           "| Code Commit Confirmation Gate | Phase 3.4 + `trellis-push` | exact Git safety checks |",
@@ -480,9 +480,11 @@
           "A same-item hit resumes the existing state",
           "`active-work-conflict` blocks unrelated code writes until the current item is completed, explicitly abandoned, or adopted into a task",
           "Unrelated read-only requests may continue without mutating the state.",
-          "python3 ./.trellis/scripts/untracked_flow.py prepare-edit --paths <exact intended paths>",
-          "Do not edit when baseline capture or workspace validation fails.",
-          "Do not edit when baseline capture, scope extension, or workspace validation fails.",
+          "A successful `begin` starts at `stage=implement`; the helper is a workflow cursor",
+          "[workflow-state:untracked_check]",
+          "[workflow-state:untracked_spec]",
+          "[workflow-state:untracked_push]",
+          "`stage=push` is only a route cursor",
           "python3 ./.trellis/scripts/task_intent.py adopt \"<title>\" --slug <slug>",
           "adoption continues through planning artifacts, Brief review, and `task.py start`, and never authorizes immediate implementation",
           "task_intent.py discard --task <current-task>"
@@ -1207,7 +1209,7 @@
           "<!-- BEGIN skill-garden patch trellis-meta-managed-workflow-source v0.6 -->",
           "<!-- BEGIN skill-garden patch trellis-meta-managed-owner-routing v0.6 -->\n## Skill Routing",
           "Do not choose implementation or checking behavior from a static platform-capability split.",
-          "| Untracked task adoption | `workflow-state:untracked`, `trellis-brainstorm`, and `task_intent.py adopt` |",
+          "| Untracked task adoption | `workflow-state:untracked*`, `trellis-brainstorm`, and `task_intent.py adopt` |",
           "| Automatic task loop and return gate | `trellis-auto-loop`",
           "| Commit/push safety and completion activation | `trellis-push` and `task_progress.py` |",
           "| Cross-session task progress discovery and recovery | `trellis-continue` owns the recovery decision, `task_progress.py` owns candidate evidence and completed-task reopen, and `task.py start` with `.trellis/scripts/common/active_task.py` owns explicit session binding |",
@@ -1236,7 +1238,7 @@
           "<!-- BEGIN skill-garden patch trellis-meta-managed-workflow-source v0.6 -->",
           "<!-- BEGIN skill-garden patch trellis-meta-managed-owner-routing v0.6 -->\n## Skill Routing",
           "Do not choose implementation or checking behavior from a static platform-capability split.",
-          "| Untracked task adoption | `workflow-state:untracked`, `trellis-brainstorm`, and `task_intent.py adopt` |",
+          "| Untracked task adoption | `workflow-state:untracked*`, `trellis-brainstorm`, and `task_intent.py adopt` |",
           "| Automatic task loop and return gate | `trellis-auto-loop`",
           "| Commit/push safety and completion activation | `trellis-push` and `task_progress.py` |",
           "| Cross-session task progress discovery and recovery | `trellis-continue` owns the recovery decision, `task_progress.py` owns candidate evidence and completed-task reopen, and `task.py start` with `.trellis/scripts/common/active_task.py` owns explicit session binding |",

package/enhancements/0.6/overrides/patches/agents/untracked-context/codex-content.md

@@ -1,2 +1,2 @@
 
-If the dispatch prompt's first line is `Untracked work: <work-id>`, run `python3 ./.trellis/scripts/untracked_flow.py status --verbose`, require the same work id, use the complete summary/scope/baseline/fingerprint/evidence/spec context supplied by the main agent, and skip any task lookup or task-artifact loading below. Do not require or invent a task path or JSONL file. Otherwise continue with the agent's normal task context rules.
+If the dispatch prompt's first line is `Untracked work: <work-id>`, run `python3 ./.trellis/scripts/untracked_flow.py status --verbose`, require the same work id, use its summary/stage plus the actual diff, relevant specs, validation context, and responsibility supplied by the main agent, and skip any task lookup or task-artifact loading below. The helper is only a workflow cursor; do not require or invent a task path, JSONL file, scope, baseline, fingerprint, or owner evidence. Otherwise continue with the agent's normal task context rules.

package/enhancements/0.6/overrides/patches/agents/untracked-context/markdown-content.md

@@ -3,6 +3,6 @@
 Resolve exactly one dispatch subject from the first prompt line:
 
 - `Active task: <path>`: use the normal task path. Read the role manifest (`implement.jsonl` for implement, `check.jsonl` for check), every listed file, then `prd.md`, optional `design.md`, and optional `implement.md`.
-- `Untracked work: <work-id>`: run `python3 ./.trellis/scripts/untracked_flow.py status --verbose`, require the same work id, and use the complete summary/scope/baseline/fingerprint/evidence/spec context supplied by the main session. Do not require or invent task artifacts or JSONL files.
+- `Untracked work: <work-id>`: run `python3 ./.trellis/scripts/untracked_flow.py status --verbose`, require the same work id, and use its summary/stage plus the actual diff, relevant specs, validation context, and responsibility supplied by the main session. The helper is only a workflow cursor; do not require or invent task artifacts, JSONL files, scope, baseline, fingerprint, or owner evidence.
 
 If hook-injected context is present, it may satisfy the task branch. The untracked branch remains prompt-driven and must still validate the helper state. If neither first-line contract is present or the resolved subject mismatches, stop and report the missing context; do not guess or switch subjects.

package/enhancements/0.6/overrides/patches/hooks/inject-workflow-state/shared-runtime/main-content.py

@@ -29,10 +29,11 @@ def main() -> int:
             )
         else:
             work_id, stage, summary = untracked
-            untracked_key = resolve_breadcrumb_key("untracked", platform, config)
+            untracked_status = "untracked" if stage == "implement" else f"untracked_{stage}"
+            untracked_key = resolve_breadcrumb_key(untracked_status, platform, config)
             breadcrumb = build_breadcrumb(
                 None,
-                "untracked",
+                untracked_status,
                 templates,
                 breadcrumb_key=untracked_key,
                 subject_label=f"Untracked work: {work_id} ({stage})",

package/enhancements/0.6/overrides/patches/hooks/inject-workflow-state/shared-runtime/untracked-helper-content.py

@@ -12,7 +12,6 @@ def _get_untracked_work(root: Path, input_data: dict) -> Optional[tuple[str, str
             root,
             input_data,
             platform=_detect_platform(input_data),
-            validate_workspace=False,
         )
     except Exception:
         return None

package/enhancements/0.6/overrides/patches/skills/trellis-meta/managed-workflow-owners/owner-routing-content.md

@@ -8,8 +8,8 @@ Stable owner categories are:
 | --- | --- |
 | Request intent and project knowledge discovery | Request Triage, `trellis-start`, and the referenced router helper |
 | Active task scope safety | Request Triage and the active-task scope guard |
-| Untracked work completion | `workflow-state:untracked`, Phase 2/3 owners, and `untracked_flow.py` |
-| Untracked task adoption | `workflow-state:untracked`, `trellis-brainstorm`, and `task_intent.py adopt` |
+| Untracked work completion | `workflow-state:untracked*`, Phase 2/3 owners, and `untracked_flow.py` |
+| Untracked task adoption | `workflow-state:untracked*`, `trellis-brainstorm`, and `task_intent.py adopt` |
 | Planning handoff and activation | `trellis-task-brief` and the task-start Brief guard |
 | Implement/check execution mode | `trellis-route` |
 | Unified quality verification | `trellis-check-all` |

package/enhancements/0.6/overrides/patches/skills/trellis-meta/managed-workflow-owners/workflow-change-map-content.md

@@ -6,7 +6,7 @@ Start from the runtime section, then move to its owner:
 | --- | --- |
 | Add or reorder a phase | Workflow Patch/source plus every affected owner handoff |
 | Change task creation or scope policy | Request Triage, task-intent helper, and the managed workflow/state Patch |
-| Change untracked completion or adoption | `workflow-state:untracked`, Phase 2/3 owners, `untracked_flow.py`, and `task_intent.py adopt` |
+| Change untracked completion or adoption | `workflow-state:untracked*`, Phase 2/3 owners, `untracked_flow.py`, and `task_intent.py adopt` |
 | Change planning handoff or activation | `trellis-task-brief` and the task-start Brief guard |
 | Change implement/check execution | `trellis-route`; Check-All remains the unified check entry |
 | Change automatic continuation | `trellis-auto-loop` and its runner action contract |

package/enhancements/0.6/overrides/patches/skills/trellis-start/no-task-routing/content.md

@@ -1 +1 @@
-- **No active task** -> first run `python3 ./.trellis/scripts/untracked_flow.py status`. On `hit`, resume the reported stage through `[workflow-state:untracked]`; on `miss`, follow the workflow `Request Triage` contract before acting. Use `task_intent.py create` for inferred complex planning, `task.py create` for explicit task planning, and ask only for material ambiguity or an independent safety boundary.
+- **No active task** -> first run `python3 ./.trellis/scripts/untracked_flow.py status`. On `hit`, resume the reported stage through the matching `[workflow-state:untracked*]` breadcrumb; on `miss`, follow the workflow `Request Triage` contract before acting. Use `task_intent.py create` for inferred complex planning, `task.py create` for explicit task planning, and ask only for material ambiguity or an independent safety boundary.

package/enhancements/0.6/overrides/patches/skills/trellis-update-spec/autonomous-evaluation/content.md

@@ -27,7 +27,7 @@ Read real evidence in this order. Do not decide from the chat summary, task titl
 
 1. For task context, the current task's `implement.jsonl` / `check.jsonl` and every file they reference.
 2. For task context, the current task's `prd.md`, `design.md`, and `implement.md`.
-3. For untracked context, `untracked_flow.py status --verbose`, including summary, scope, baseline/current fingerprint, focused validation, and Check-All evidence.
+3. For untracked context, `untracked_flow.py status --verbose` for work id, summary, and stage; obtain validation evidence from the current Check-All result and current change set rather than the cursor helper.
 4. The final Check-All conclusion and its actual validation evidence.
 5. The current subject's actual diff, source code, tests, and commit evidence.
 6. Existing specs and their indexes returned by `spec_router.py`.
@@ -56,6 +56,6 @@ When applicable, also validate indexes/links, code signatures, or project-specif
 - Interactive: after a passed Check-All stop, when the user says “下一步”, “继续”, `next`, `continue`, or an equivalent continuation intent, run this skill. A `no-op` or `written` result must load `trellis-push` in the same turn and present its single confirmation plan. A `needs-review` result stops and must not generate a Push plan.
 - Interactive direct Git: when the latest user message that triggered the current completion chain explicitly requests an ordinary push or a user-initiated `commit-only`, use that request only as conditional continuation after a strictly passed Check-All. After the existing standard Check-All report is shown, run this skill in the same turn when no currently valid `spec_update_result` exists. Only `no-op` or `written` may proceed to `trellis-push`; `needs-review` stops. Do not infer this intent from history, summaries, dirty state, or an auto-loop internal `commit-only`.
 - Validated auto-loop: for `no-op` or `written`, execute `record --action run_spec_update --result ok` and immediately run `next`. For `needs-review`, execute `record --action run_spec_update --result blocked --failure-type spec-needs-review`; never disguise it as `no-op`.
-- Untracked: after the final spec diff stabilizes, run `untracked_flow.py record-spec --result <no-op|written|needs-review> --summary <summary>`. Only `no-op` or `written` may advance to `push`; `needs-review` stays at `spec`. Any later edit outside this skill's allowed `.trellis/spec/**` batch must return through `prepare-edit` and invalidate downstream evidence.
+- Untracked: keep the cursor at `spec` for `needs-review`. For `no-op` or `written`, run `untracked_flow.py advance --stage push`. Any later product edit returns the cursor to `implement`; the helper does not validate or preserve owner evidence.
 
 Do not ask again or rerun when a currently valid `no-op` or `written` result already exists. Re-evaluate after the actual diff, Check-All conclusion, or the user's spec intent changes.

package/enhancements/0.6/overrides/patches/workflow/hub/content.md

@@ -12,8 +12,8 @@ Complete contracts live in the owning phase, workflow state, skill, hook, or hel
 | Project Knowledge Discovery | `Request Triage` | `spec_router.py` |
 | Flower Update Confirmation | SessionStart update context + Flower CLI | update hook / `self-update` arguments |
 | Active Task Scope Guard | `Request Triage` | `task_intent.py` scope safety |
-| Untracked Work Completion Chain | `workflow-state:untracked` + Phase 2/3 owners | `untracked_flow.py` |
-| Untracked Task Adoption | `workflow-state:untracked` + `trellis-brainstorm` | `task_intent.py adopt` |
+| Untracked Work Completion Chain | `workflow-state:untracked*` + Phase 2/3 owners | `untracked_flow.py` |
+| Untracked Task Adoption | `workflow-state:untracked*` + `trellis-brainstorm` | `task_intent.py adopt` |
 | Routing Gate | Phase 2 + `trellis-route` | `route_state.py` |
 | Auto-Loop Return Gate | `trellis-check-all` + `trellis-auto-loop` | `auto_loop.py record/next` |
 | Interactive Post-Check Stop Gate | Phase 2.2 + `trellis-check-all` | current Check-All evidence |

package/enhancements/0.6/overrides/patches/workflow/phase-ownership/phase-2-check-content.md

@@ -2,7 +2,7 @@
 
 Run `trellis-route(target=check)`, then execute the unified `trellis-check-all` entry using the validated inline/subagent route.
 
-For untracked work, route reads only the personal pref helper and the dispatch prompt starts with `Untracked work: <work-id>` plus complete state/spec context. After Check-All, record its result through `untracked_flow.py record-check`; only a current valid pass may advance the work to `spec`.
+For untracked work, route reads only the personal pref helper and the dispatch prompt starts with `Untracked work: <work-id>` plus the work summary, current stage, actual diff, relevant specs, and validation context. Findings or new edits return the cursor to `implement`; a strict pass advances to `spec` only when the existing interactive/direct-Git disposition continues the completion chain.
 
 Before interactive Check-All begins, run `python3 ./.trellis/scripts/pre_check_state.py clear`. A missing, subject-mismatched, or already-cleared preference is a no-op; a damaged runtime is reported diagnostically but safely defaults to checking.
 

package/enhancements/0.6/overrides/patches/workflow/phase-ownership/phase-2-implement-content.md

@@ -7,13 +7,13 @@ Follow the validated route result:
 - `inline`: load `trellis-before-dev`, read the active task artifacts and referenced context, then implement and run focused verification.
 - `subagent`: dispatch the selected implement agent with `Active task: <task path>` as the first prompt line; the agent implements directly and must not recursively dispatch implement/check agents.
 
-For untracked work, route reads only the personal pref helper and never creates task-scoped route decisions. Inline loads relevant project specs from `spec_router.py`; subagent dispatch starts with `Untracked work: <work-id>` and includes the helper's complete state summary instead of task artifacts. Run `untracked_flow.py prepare-edit --paths <exact intended paths>` before every write batch.
+For untracked work, route reads only the personal pref helper and never creates task-scoped route decisions. Inline loads relevant project specs from `spec_router.py`; subagent dispatch starts with `Untracked work: <work-id>` and includes the work summary, current stage, actual diff/spec context, and this turn's responsibility instead of task artifacts. If implementation resumes from a downstream stage, set the cursor to `implement` before editing.
 
 Route preference recovery, fallback choices, and runtime evidence belong to `trellis-route`; do not reproduce them here.
 
 After implementation and focused verification, resolve the next action in this order:
 
-For untracked work, first record the focused validation through `untracked_flow.py record-validation`; before entering Check-All, advance to `check`. A failed or partial validation remains in `implement` until the owner evidence permits advancement.
+For untracked work, focused validation remains owned by the implementation path. When it is complete, advance the cursor to `check`; a failed or partial validation stays at `implement`.
 
 1. A validated auto-loop outstanding action wins; continue to its requested Check-All action without consulting interactive hold state.
 2. If the latest user message explicitly requests checking, continuation, commit, or deployment, run `python3 ./.trellis/scripts/pre_check_state.py clear` and enter Phase 2.2 in the same turn.

package/enhancements/0.6/overrides/patches/workflow/phase-ownership/phase-3-commit-content.md

@@ -2,7 +2,7 @@
 
 Load `trellis-push`. It owns dirty-file classification, exact file/message planning, one-shot confirmation, Git safety checks, ordinary commit + push, and current-task progress synchronization.
 
-For untracked work, the current state baseline/scope/evidence owns attribution and there is no task progress commit. After every confirmed repository action succeeds and the final workspace evidence is recorded, clear the untracked state with `--reason completed`; partial or failed Git execution keeps the state for recovery.
+For untracked work, `stage=push` identifies the next owner and there is no task progress commit. `trellis-push` derives attribution from the current request, work summary, and actual diff. After every confirmed repository action succeeds, clear the untracked state with `--reason completed`; partial or failed Git execution keeps the cursor for recovery.
 
 Ordinary mode defaults to commit and push. Commit-only is allowed only when the user explicitly requests a local commit or a validated auto-loop supplies its scoped preauthorization.
 

package/enhancements/0.6/overrides/patches/workflow/phase-ownership/phase-3-update-spec-content.md

@@ -8,4 +8,4 @@ Load `trellis-update-spec` and let it decide whether the task produced executabl
 
 Do not ask a separate generic “update spec?” question before invoking the skill.
 
-For untracked work, evaluate the current state/diff/spec evidence without inventing task artifacts, record the result through `untracked_flow.py record-spec`, and advance to `push` only for `no-op` or `written`.
+For untracked work, evaluate the actual diff and relevant specs without inventing task artifacts. Keep the cursor at `spec` for `needs-review`; advance it to `push` only for `no-op` or `written`.

package/enhancements/0.6/overrides/patches/workflow/runtime-contract-reference/customizing-trellis-managed-baseline.md

@@ -60,7 +60,7 @@ Supported events: `after_create / after_start / after_finish / after_archive`. N
 ### Full contract
 
 <!-- BEGIN skill-garden patch workflow-runtime-contract-reference v0.6 -->
-For the workflow state machine's runtime contract, the authoritative runtime inputs are the installed per-turn hook parser and the `[workflow-state:*]` tags in this file. This Flower variant uses fixed pseudo-status tag names `no_task`, `untracked`, and `missing_task`; hook diagnostic source types such as `session` or `session-fallback` must not become workflow-state tag names.
+For the workflow state machine's runtime contract, the authoritative runtime inputs are the installed per-turn hook parser and the `[workflow-state:*]` tags in this file. This Flower variant uses fixed pseudo-status tag names `no_task`, `untracked`, `untracked_check`, `untracked_spec`, `untracked_push`, and `missing_task`; hook diagnostic source types such as `session` or `session-fallback` must not become workflow-state tag names.
 
 - Installed `<platform>/hooks/inject-workflow-state.py` copies — parse this workflow and emit the current breadcrumb for platforms with a per-turn hook.
 - `.trellis/spec/` project specs, when present — project-local runtime contract notes and invariants.

package/enhancements/0.6/overrides/patches/workflow/runtime-contract-reference/runtime-reference-content.md

@@ -1,4 +1,4 @@
-For the workflow state machine's runtime contract, the authoritative runtime inputs are the installed per-turn hook parser and the `[workflow-state:*]` tags in this file. This Flower variant uses fixed pseudo-status tag names `no_task`, `untracked`, and `missing_task`; hook diagnostic source types such as `session` or `session-fallback` must not become workflow-state tag names.
+For the workflow state machine's runtime contract, the authoritative runtime inputs are the installed per-turn hook parser and the `[workflow-state:*]` tags in this file. This Flower variant uses fixed pseudo-status tag names `no_task`, `untracked`, `untracked_check`, `untracked_spec`, `untracked_push`, and `missing_task`; hook diagnostic source types such as `session` or `session-fallback` must not become workflow-state tag names.
 
 - Installed `<platform>/hooks/inject-workflow-state.py` copies — parse this workflow and emit the current breadcrumb for platforms with a per-turn hook.
 - `.trellis/spec/` project specs, when present — project-local runtime contract notes and invariants.

package/enhancements/0.6/overrides/patches/workflow/runtime-contract-reference/state-contract-comment-content.md

@@ -21,7 +21,7 @@
 
   TAG ↔ PHASE scoping:
     [workflow-state:no_task]      → no active task; before Phase 1
-    [workflow-state:untracked]    → current-session direct-edit completion chain
+    [workflow-state:untracked*]   → current-session direct-edit stage cursor
     [workflow-state:missing_task]   → missing active-task directory recovery
     [workflow-state:planning]     → all of Phase 1 (status='planning')
     [workflow-state:planning-inline] → Codex inline variant of Phase 1
@@ -41,6 +41,6 @@
       in Skill-Garden managed mode, edit the owning Patch and run source sync,
       compiled-target checks, and dogfood update
     - Runtime pseudo-status names are fixed. This Flower variant defines
-      `no_task`, `untracked`, and `missing_task`; hook diagnostic source types must not be
+      `no_task`, the `untracked*` stage family, and `missing_task`; hook diagnostic source types must not be
       appended to workflow-state tag names.
 -->

package/enhancements/0.6/overrides/patches/workflow/state-no-task/content.md

@@ -2,5 +2,5 @@ No active task. Infer the current request intent before acting.
 Repair intent alone is not a no-task switch; inspect unknown scope and reclassify before edits. Only an explicit current-request workflow instruction such as `直接做` / `不要任务` may override automatic `task_plan`.
 For non-trivial project work, follow the `Request Triage` Project Knowledge Discovery contract before routing the action. Load a Trellis capability directly only when the user explicitly names it or the request exactly matches that capability; route project-specific workflow actions through the matched SOP instead of keyword-mapping a general release/publish request to `trellis-release`.
 Handle `discuss` and `inspect` silently. For non-destructive `direct_edit`, state once that work will use the session-scoped untracked flow, then run `python3 ./.trellis/scripts/untracked_flow.py begin --summary "<bounded work summary>" --source <inferred|user-explicit>`.
-A same-item hit resumes the existing state; `active-work-conflict` blocks unrelated code writes until the current item is completed, explicitly abandoned, or adopted into a task. Unrelated read-only requests may continue without mutating the state. After a successful `begin`, run `python3 ./.trellis/scripts/untracked_flow.py prepare-edit --paths <exact intended paths>` before the first file write. Do not edit when baseline capture or workspace validation fails.
+A same-item hit resumes the existing state; `active-work-conflict` blocks unrelated code writes until the current item is completed, explicitly abandoned, or adopted into a task. Unrelated read-only requests may continue without mutating the state. A successful `begin` starts at `stage=implement`; the helper is a workflow cursor and does not gate edits on file scope, Git state, or owner evidence.
 For high-confidence complex implementation, create an auto-routed planning task through `task_intent.py create`, show one non-blocking switch hint, and enter `trellis-brainstorm`. Ask only for material ambiguity or independent safety gates.

package/enhancements/0.6/overrides/patches/workflow/state-untracked/content.md

@@ -1,8 +1,20 @@
 [workflow-state:untracked]
-One current-session untracked work item is active. Run `python3 ./.trellis/scripts/untracked_flow.py status` and continue its reported stage; do not create task artifacts or task-scoped route decisions.
-At `inspect` or `implement`, enter Phase 2.1 with `trellis-route(target=implement)` and run `prepare-edit --paths <exact paths>` before every write batch. Do not edit when baseline capture, scope extension, or workspace validation fails. Use `record-validation` for focused evidence, then advance to `check` only when it passes.
-At `check`, enter Phase 2.2 with `trellis-route(target=check)`; use `record-check` for the Check-All result and advance to `spec` only on a current valid pass.
-At `spec`, run `trellis-update-spec`; use `record-spec` for `no-op`, `written`, or `needs-review`, and advance to `push` only for the first two results.
-At `push`, load `trellis-push`; clear with `--reason completed` only after all confirmed Git actions succeed. New edits return through `prepare-edit`, which invalidates downstream evidence.
-A different implementation request is blocked by the single-active-work guard until this item is completed, explicitly abandoned, or adopted through `python3 ./.trellis/scripts/task_intent.py adopt "<title>" --slug <slug>`; adoption continues through planning artifacts, Brief review, and `task.py start`, and never authorizes immediate implementation. Unrelated read-only requests may continue without changing state.
+The current-session untracked work item is at `implement`. Run `python3 ./.trellis/scripts/untracked_flow.py status`, then enter Phase 2.1 through `trellis-route(target=implement)` without creating task artifacts or task-scoped route decisions.
+The helper is only a workflow cursor: it does not validate file scope, Git state, focused validation, or owner evidence. After focused validation is complete, run `python3 ./.trellis/scripts/untracked_flow.py advance --stage check` and continue through the normal Phase 2.1 completion contract.
+A different implementation request remains blocked by the single-active-work guard until this item is completed, explicitly abandoned, or adopted through `python3 ./.trellis/scripts/task_intent.py adopt "<title>" --slug <slug>`; adoption continues through planning artifacts, Brief review, and `task.py start`, and never authorizes immediate implementation. Unrelated read-only requests may continue without changing state.
 [/workflow-state:untracked]
+
+[workflow-state:untracked_check]
+The current-session untracked work item is at `check`. Run `python3 ./.trellis/scripts/untracked_flow.py status`, then enter Phase 2.2 through `trellis-route(target=check)` and execute `trellis-check-all`.
+If Check-All reports findings or a new edit is needed, run `python3 ./.trellis/scripts/untracked_flow.py advance --stage implement` before returning to implementation. On a strict pass, keep `check` while the interactive stop gate waits; only a same-turn direct Git continuation or a later explicit continuation advances to `spec`.
+[/workflow-state:untracked_check]
+
+[workflow-state:untracked_spec]
+The current-session untracked work item is at `spec`. Run `python3 ./.trellis/scripts/untracked_flow.py status`, then load `trellis-update-spec`.
+Keep `spec` for `needs-review`. For `no-op` or `written`, run `python3 ./.trellis/scripts/untracked_flow.py advance --stage push`. The helper records only the next owner; Update-Spec owns its evidence and validation.
+[/workflow-state:untracked_spec]
+
+[workflow-state:untracked_push]
+The current-session untracked work item is ready to enter Push. Run `python3 ./.trellis/scripts/untracked_flow.py status`, require `stage=push`, and load `trellis-push`.
+`stage=push` is only a route cursor: it is not a Git plan, user confirmation, or proof that Push already ran. `trellis-push` owns the exact plan, confirmation, Git safety checks, and execution. Clear with `--reason completed` only after every confirmed Git action succeeds; new edits first return the cursor to `implement`.
+[/workflow-state:untracked_push]

package/enhancements/0.6/scripts/pre_check_state.py

@@ -19,7 +19,7 @@ PREFERENCE_KEY = "pre_check_preference"
 PREFERENCE_VERSION = 2
 LEGACY_PREFERENCE_VERSION = 1
 VALID_SOURCES = {"user-explicit", "follow-up-edit"}
-UNTRACKED_STATE_VERSION = 1
+UNTRACKED_STATE_VERSIONS = {1, 2}
 UNTRACKED_STAGES = {"inspect", "implement", "check", "spec", "push"}
 SESSION_START_HINT = "Pre-check: deferred for current work; latest user intent may override."
 
@@ -137,7 +137,7 @@ def _runtime_scope(
     if isinstance(untracked, dict):
         work_id = untracked.get("id")
         if not (
-            untracked.get("version") == UNTRACKED_STATE_VERSION
+            untracked.get("version") in UNTRACKED_STATE_VERSIONS
             and isinstance(work_id, str)
             and work_id.strip()
             and untracked.get("stage") in UNTRACKED_STAGES

package/enhancements/0.6/scripts/task_intent.py

@@ -243,7 +243,7 @@ def _rollback_adoption(
 
 
 def adopt_untracked_work(args: argparse.Namespace) -> dict:
-    """把当前 untracked diff 和证据原地接管到 planning task。
+    """把当前 untracked diff 和流程游标原地接管到 planning task。
 
     Args:
         args: adopt 子命令参数。
@@ -256,7 +256,7 @@ def adopt_untracked_work(args: argparse.Namespace) -> dict:
     if not context_key:
         raise IntentTaskError("no-session-context", "无法解析当前 AI session")
     try:
-        untracked = read_untracked_state(repo_root, validate_workspace=True)
+        untracked = read_untracked_state(repo_root)
     except Exception as exc:
         reason = getattr(exc, "reason", "untracked-state-invalid")
         raise IntentTaskError(reason, str(exc)) from exc
@@ -267,6 +267,7 @@ def adopt_untracked_work(args: argparse.Namespace) -> dict:
     raw_state = session_before.get(STATE_KEY)
     if not isinstance(raw_state, dict) or raw_state.get("id") != state["id"]:
         raise IntentTaskError("untracked-state-mismatch", "纳管前 untracked 状态已变化")
+    baseline = capture_git_baseline(repo_root)
 
     task_ref = ""
     task_dir: Path | None = None
@@ -281,11 +282,9 @@ def adopt_untracked_work(args: argparse.Namespace) -> dict:
             "createdAt": _utc_now(),
             "contextKey": context_key,
             "untrackedWorkId": state["id"],
-            "implementationStarted": state["stage"] != "inspect",
-            "baseline": state.get("baseline"),
+            "implementationStarted": True,
+            "baseline": baseline,
             "adoptedStage": state["stage"],
-            "workspaceFingerprint": state.get("workspaceFingerprint"),
-            "evidence": state.get("evidence", {}),
         }
         task_data["meta"] = meta
         _write_json_atomic(task_json, task_data)
@@ -322,7 +321,7 @@ def adopt_untracked_work(args: argparse.Namespace) -> dict:
         "task": task_ref,
         "workId": state["id"],
         "adoptedStage": state["stage"],
-        "implementationStarted": state["stage"] != "inspect",
+        "implementationStarted": True,
     }