loopx-project · huangruiteng · Sep 20, 2026 · Sep 20, 2026
@@ -138,6 +138,7 @@ export function ContextDrawer({ agents, attentionHistory = [], onSelectAttention
   const lastAuthoritativeSubagentConfigurationRef = useRef<WorkspaceGoalSubagentConfiguration | null>(null);
   const verifiedSubagentBaselineRef = useRef<WorkspaceGoalSubagentConfiguration | null>(null);
   const [todoAgentId, setTodoAgentId] = useState(agents.find((agent) => agent.available)?.agentId ?? "codex");
+  const [todoPriority, setTodoPriority] = useState("");
   const [todoResumeWhen, setTodoResumeWhen] = useState("");
   const closeRef = useRef<HTMLButtonElement>(null);
   const drawerRef = useRef<HTMLDivElement>(null);
@@ -607,6 +608,21 @@ export function ContextDrawer({ agents, attentionHistory = [], onSelectAttention
                       summary: t("drawer.reassignSummary", { task: selection.item.text }),
                     })} type="button">{t("timeline.review")}</button>
                   </label>
+                  <label className="personal-inline-agent-select">{t("drawer.taskPriority")}
+                    <select aria-label={t("drawer.taskPriority")} value={todoPriority} onChange={(event) => setTodoPriority(event.target.value)}>
+                      <option value="">{t("drawer.taskPriorityChoose")}</option>
+                      {["P0", "P1", "P2", "P3", "P4"].map((priority) => <option key={priority} value={priority}>{priority}</option>)}
+                      <option value="clear">{t("drawer.taskPriorityClear")}</option>
+                    </select>
+                    <button className="personal-secondary-action" disabled={!todoPriority} onClick={() => void callbacks.onPreviewAction?.({
+                      actionKind: "todo.update",
+                      context: {goal_id: selection.item.goalId, kind: "todo", todo_id: selection.item.todoId},
+                      idempotencyKey: `workspace-todo-${selection.item.todoId}-priority-${todoPriority}-${Date.now().toString(36)}`,
+                      normalizedParameters: {goal_id: selection.item.goalId, todo_id: selection.item.todoId, operation: "edit",
+                        ...(todoPriority === "clear" ? {clear_priority: true} : {priority: todoPriority})},
+                      summary: `${t("drawer.taskPriority")}: ${todoPriority === "clear" ? t("drawer.taskPriorityClear") : todoPriority}`,
+                    })} type="button">{t("timeline.review")}</button>
+                  </label>
                   <strong>{t("drawer.taskDeferUntil")}</strong>
                   <label className="personal-inline-agent-select personal-inline-resume-when">{t("drawer.taskDeferUntil")}
                     <input

@@ -140,7 +140,7 @@ export function GoalTasksView({
     .filter((todo) => todo.goalId === goal.goalId)
     .map((todo) => ({ ...todo, goalTitle: goal.title }));
   const priorityRank = (todo: WorkspaceGoal["agentTodos"][number]) =>
-    todo.priority === "P0" ? 0 : todo.priority === "P1" ? 1 : todo.priority === "P2" ? 2 : 3;
+    todo.priority === "P0" ? 0 : todo.priority === "P1" ? 1 : todo.priority === "P2" ? 2 : todo.priority === "P3" ? 3 : todo.priority === "P4" ? 4 : 50;
   const agentLanes = useMemo(() => {
     const lanes = new Map((goal.agentLanes ?? []).map((lane) => [lane.agentId, lane]));
     for (const todo of goal.agentTodos) {

@@ -350,6 +350,9 @@ const en = {
   "drawer.taskDeferPlaceholder": "For example, resume_at:2026-09-14T09:00:00+08:00",
   "drawer.taskDeferReview": "Review defer",
   "drawer.taskDeferSupported": "Supports todo_done, pr_merged, capacity_available, and timezone-aware resume_at conditions.",
+  "drawer.taskPriority": "Priority",
+  "drawer.taskPriorityChoose": "Choose priority",
+  "drawer.taskPriorityClear": "No priority",
   "drawer.taskDeferUntil": "Defer until",
   "drawer.taskDetails": "Todo details",
   "drawer.taskInfo": "Task information",
@@ -1432,6 +1435,9 @@ const zhCN: Record<WorkspaceMessageKey, string> = {
   "drawer.taskDeferPlaceholder": "例如 resume_at:2026-09-14T09:00:00+08:00",
   "drawer.taskDeferReview": "检查暂缓",
   "drawer.taskDeferSupported": "支持 todo_done、pr_merged、capacity_available 与带时区的 resume_at 条件。",
+  "drawer.taskPriority": "优先级",
+  "drawer.taskPriorityChoose": "选择优先级",
+  "drawer.taskPriorityClear": "不设优先级",
   "drawer.taskDeferUntil": "暂缓至",
   "drawer.taskDetails": "Todo 详情",
   "drawer.taskInfo": "任务信息",

@@ -2759,6 +2759,15 @@ The typed TypeScript `projection_delivery` union also owns the distinction
 between mutation intent (`pending`/`not_required`) and provider readback
 (`delivered`/`current`); unknown states fail closed before acknowledgement.
 
+Priority intent now follows the same admitted create/update transaction on File,
+SQLite and PostgreSQL. Explicit set/clear, omission and conflicting legacy text
+are resolved by `todos/priority.ts`; Python reads share the generated grammar.
+Markdown remains compatible display, while native records retain matching
+priority/title. CLI and reviewed Chat edits preserve CAS and historical retry
+identity. Real backend readback and a disposable clone of a long-lived local
+Goal qualify this bounded change. See the [caller contract](../../project-agent-todo-contract.md#priority-intent).
+This does not change provider defaults or close the remaining promotion gates.
+
 Presentation is canonical at the projection layer, not in the domain record.
 `source_section` and `index` are the v0 wire shape's display coordinates, while
 native records derive the same display section from role/archive state and use

@@ -2188,6 +2188,13 @@ Python 生产 caller 直接从 `todos/todo_semantics.py` 导入；`todos/project
 mutation intent（`pending`/`not_required`）与 provider readback（`delivered`/`current`）；
 未知状态在 acknowledgement 之前 fail closed。
 
+优先级意图现接入 File、SQLite、PostgreSQL 既有的准入 create/update 事务。
+显式设置/清除、参数缺省及与旧文字前缀的冲突由 `todos/priority.ts` 处理，Python
+读取共享生成的语法。Markdown 保留兼容展示，native record 保存一致的 priority/title。
+CLI 与经过审阅的 Chat 编辑保留 CAS 和历史重试身份。真实后端回读及长期本地 Goal 的
+一次性隔离副本验证这条边界，见[调用合同](../../project-agent-todo-contract.md#priority-intent)。
+这不改变 provider 默认，也不关闭其余 promotion 门禁。
+
 展示语义属于 projection 层，而不是 domain record。`source_section` 与 `index` 是 v0
 wire shape 的展示坐标；native record 根据 role/archive state 推导相同的展示 section，
 并以时间戳和 Todo identity 做确定性回退，不制造假的持久 index。因此即使 wire shape

@@ -58,12 +58,23 @@ remaining lifecycle writers have migrated.
 
 The same stage also removes duplicated Python read policy around that boundary.
 Task-class resolution, title-aware actionability, dependency readiness, agent
-eligibility, priority ordering, and canonical Todo read records now have one
-Python semantic owner while TypeScript remains the transaction owner. The old
+eligibility and canonical Todo read records have one Python semantic owner
+while TypeScript remains the transaction owner. Priority authoring and ordering
+now share `todos/priority.ts`; Python compatibility readers consume its generated
+vocabulary and legacy grammar instead of maintaining independent patterns. The old
 projection module is an import-only compatibility facade. This keeps the
 replacement-first rule intact: compatibility remains available, but it cannot
 silently become a second semantic implementation.
 
+The priority-intent slice connects CLI add/update/clear, reviewed Chat edits and
+the Dashboard selector to the existing typed Todo transaction. Text-only edits
+preserve priority; conflicting declarations fail before writes. P3/P4 ordering,
+legacy decorated labels and successor inheritance share the same owner. See the
+[caller contract](../../project-agent-todo-contract.md#priority-intent). Real CLI
+File/SQLite readback, isolated PostgreSQL and the shared complex fixture qualify
+this boundary. It retires duplicate priority knowledge, not the remaining T1/T3
+callers, Python compatibility IO or the D1–D3 default-cutover gates.
+
 Native update now composes `todos/public_update.ts` for a bounded nonterminal
 planning intent (status, evidence/reason, resume/clear and successor links),
 against the same complete canonical head used for authority checks and CAS.

@@ -37,11 +37,19 @@ fixture 覆盖。这是一个已完成的交付阶段，不代表 Markdown 晋
 lifecycle writer 已全部迁移。
 
 同一阶段也删除了该边界周围重复的 Python read policy。task-class 解析、识别 title
-的 actionable 判断、依赖就绪、Agent eligibility、priority 排序和 canonical Todo
-read record 现在只有一个 Python 语义 owner，而 TypeScript 仍是事务 owner。旧
+的 actionable 判断、依赖就绪、Agent eligibility 和 canonical Todo read record
+只有一个 Python 语义 owner，TypeScript 仍是事务 owner。优先级写入意图与排序现共用
+`todos/priority.ts`；Python 兼容读取使用生成的词表和旧格式语法，不再独立维护模式。旧
 projection 模块只保留 import-only 兼容 facade。这样继续遵守 replacement-first：
 兼容路径仍可用，但不能静默形成第二份语义实现。
 
+优先级意图这一批将 CLI add/update/clear、经过审阅的 Chat 编辑和 Dashboard 选择器
+接入既有 typed Todo 事务。仅改文字保留优先级，冲突声明在写入前拒绝。P3/P4 排序、
+历史装饰标签和 successor 继承使用同一 owner。见
+[调用合同](../../project-agent-todo-contract.md#priority-intent)。真实 CLI File/SQLite
+回读、隔离 PostgreSQL 和共享复杂 fixture 验证这条边界。本批删除重复的优先级知识，
+不代表其余 T1/T3 caller、Python 兼容 IO 或 D1–D3 默认切换门禁已经完成。
+
 Native update 通过 `todos/public_update.ts` 组合非终态 planning intent，在同一份
 完整 canonical head 上校验权限并 CAS。独立 intent 命名空间保留 text/note 限制
 及旧回执指纹。v2 transport 携带 lifecycle grant、authority reason、registry 来源

@@ -27,6 +27,45 @@ continue it; when it needs to fan out, split, supersede, or create successor
 work, it writes the new task through the LoopX todo lifecycle and lets the board
 sync catch up.
 
+## Priority intent
+
+Priority is an explicit scheduling field, with `P0` highest and `P4` lowest.
+Create or edit it independently of the task description:
+
+```bash
+loopx todo add --goal-id <goal-id> --role agent --priority P1 --text '<agent action>'
+loopx todo update --goal-id <goal-id> --todo-id <todo-id> --priority P3
+loopx todo update --goal-id <goal-id> --todo-id <todo-id> --clear-priority
+```
+
+Pass the registered `--agent-id` on updates when the Goal has multiple Agents.
+The existing authoring, ownership, lease and review requirements still apply.
+Chat `todo.create` accepts `priority`; reviewed `todo.update` edits accept
+`priority` or `clear_priority`. The task management panel uses that reviewed
+update path for its priority selector. Priority never grants eligibility,
+capabilities, a claim or quota; missing priority is allowed and sorts after P4.
+The generated agent authoring hint shows an explicit P1 example, not a global
+default or a prerequisite for same-turn binding.
+
+Omitting priority from an ordinary text edit preserves the current priority.
+Use `--clear-priority` to remove it. Legacy `--text '[P2] Task'` remains supported;
+a supplied prefix edits priority for old callers. An explicit parameter and a
+conflicting prefix are rejected before writing, as are simultaneous set/clear
+instructions. Words such as `P0` inside ordinary prose have no scheduling meaning.
+Historical decorated prefixes such as `[P2-review]` read as P2; an edit renders
+the normalized `[P2]` prefix. P3/P4 participate in ordering and repair suggestions
+without being silently promoted to P1. Existing repair-suggestion and historical
+event-replay defaults retain their own contracts; they do not impose a default
+on new unprioritized Todos.
+
+`todos/priority.ts` owns authoring intent and ordering. The generated coordination
+contract supplies its vocabulary and legacy grammar to Python read adapters.
+Markdown keeps the compatible `[Pn] description` display; native authority records
+also carry canonical priority/title. File, SQLite and PostgreSQL updates use the
+existing admitted head, CAS and operation receipt. A retry retains the original
+intent identity and returns its historical result; changing priority under the
+same operation ID is a conflict. No provider promotion or receipt rewrite occurs.
+
 ## Write Contract
 
 For a caller-owned runtime that already registered its Goal/Agent, generate the

@@ -1637,6 +1637,16 @@ export const typedActionsScenario = {
       await taskRow.click();
       taskManagement = page.locator("details.personal-task-management");
       await taskManagement.locator("summary").click();
+      await page.getByLabel("优先级", {exact: true}).selectOption("P4");
+      await page.screenshot({path: resolve(outputDir, "todo-priority-edit.png"), fullPage: false, animations: "disabled"});
+      await taskManagement.locator("label", {has: page.getByLabel("优先级", {exact: true})}).getByRole("button").click();
+      await page.getByText("确认执行").waitFor({state: "visible"});
+      const priorityEdit = api.actionPreviews.findLast(preview => preview.action_kind === "todo.update" && preview.normalized_parameters.priority === "P4");
+      if (!priorityEdit || priorityEdit.normalized_parameters.text !== undefined) throw new Error("Priority edit must be structured, without a text rewrite");
+      await page.getByRole("button", {name: "关闭", exact: true}).click();
+      await taskRow.click();
+      taskManagement = page.locator("details.personal-task-management");
+      await taskManagement.locator("summary").click();
       await page.getByLabel("Todo 暂缓恢复条件").fill("pr_merged:huangruiteng/loopx#3399");
       await page.screenshot({ path: resolve(outputDir, "todo-defer-resume-condition.png"), fullPage: false, animations: "disabled" });
       await taskManagement.locator(".personal-inline-resume-when").getByRole("button", { name: "检查暂缓" }).click();

@@ -256,6 +256,7 @@ def main() -> int:
         assert approved_quota["todo_write_hint"]["agent_todo_command_template"].startswith(
             f"loopx todo add --goal-id {GOAL_ID} --role agent "
         ), approved_quota
+        assert "--priority P1" in approved_quota["todo_write_hint"]["agent_todo_command_template"], approved_quota
 
         packet = run_cli(root, registry_path, "review-packet", "--goal-id", GOAL_ID, "--scan-root", str(project))
         assert packet["ok"] is True, packet

@@ -541,13 +541,15 @@ def _add_review_todo(
     registry_path: Path,
     goal_id: str,
     text: str,
+    priority: str | None = None,
     dry_run: bool,
 ) -> dict[str, Any]:
     return add_goal_todo(
         registry_path=registry_path,
         goal_id=goal_id,
         role="agent",
         text=_normalize_todo_text(text),
+        priority=priority,
         task_class="advancement_task",
         action_kind=CHAT_TODO_ACTION_KIND,
         dry_run=dry_run,
@@ -559,11 +561,13 @@ def build_todo_review_preview(
     registry_path: Path,
     goal_id: str,
     text: str,
+    priority: str | None = None,
 ) -> dict[str, Any]:
     payload = _add_review_todo(
         registry_path=registry_path,
         goal_id=goal_id,
         text=text,
+        priority=priority,
         dry_run=True,
     )
     compact = _compact_todo_payload(payload, applied=False)
@@ -576,12 +580,14 @@ def apply_todo_review_preview(
     registry_path: Path,
     goal_id: str,
     text: str,
+    priority: str | None = None,
     preview_id: str,
 ) -> dict[str, Any]:
     current_preview = _add_review_todo(
         registry_path=registry_path,
         goal_id=goal_id,
         text=text,
+        priority=priority,
         dry_run=True,
     )
     if not preview_id or preview_id != _todo_preview_fingerprint(current_preview):
@@ -596,6 +602,7 @@ def apply_todo_review_preview(
         registry_path=registry_path,
         goal_id=goal_id,
         text=text,
+        priority=priority,
         dry_run=False,
     )
     compact = _compact_todo_payload(applied, applied=True)

@@ -218,6 +218,7 @@ def _normalize(
                 allowed={
                     "goal_id",
                     "text",
+                    "priority",
                     "agent_id",
                     "endpoint_id",
                     "start_execution",
@@ -229,6 +230,8 @@ def _normalize(
                 "goal_id": goal_id,
                 "text": _text(values.get("text"), field="text", limit=400),
             }
+            if "priority" in values:
+                result["priority"] = values["priority"]
             if values.get("endpoint_id"):
                 endpoint_id = _opaque(values.get("endpoint_id"), field="endpoint_id")
                 result["endpoint_id"] = endpoint_id
@@ -261,6 +264,8 @@ def _normalize(
                     "goal_id",
                     "todo_id",
                     "text",
+                    "priority",
+                    "clear_priority",
                     "status",
                     "note",
                     "agent_id",
@@ -301,6 +306,8 @@ def _normalize(
                 result["status"] = status
             if values.get("note"):
                 result["note"] = _text(values["note"], field="note", limit=600)
+            if "priority" in values:
+                result["priority"] = values["priority"]
             if values.get("endpoint_id"):
                 endpoint_id = _opaque(values["endpoint_id"], field="endpoint_id")
                 result["endpoint_id"] = endpoint_id
@@ -322,6 +329,12 @@ def _normalize(
                 if not isinstance(values["no_followup"], bool):
                     raise ValueError("no_followup must be true or false")
                 result["no_followup"] = values["no_followup"]
+            if "clear_priority" in values:
+                if not isinstance(values["clear_priority"], bool):
+                    raise ValueError("clear_priority must be true or false")
+                result["clear_priority"] = values["clear_priority"]
+            if operation != "edit" and ("priority" in result or result.get("clear_priority")):
+                raise ValueError("priority changes require todo.update operation=edit")
             required_by_operation = {
                 "reassign": "agent_id",
                 "defer": "resume_when",
@@ -333,7 +346,7 @@ def _normalize(
             if operation == "block" and not result.get("note"):
                 raise ValueError("todo.update block requires note")
             if operation == "edit" and len(result) == 3:
-                raise ValueError("todo.update requires text, status, or note")
+                raise ValueError("todo.update requires text, priority, status, or note")
             return result
         if action_kind == "run.correct":
             values = self._allowed_parameters(