emrg: a task record names its agent with config.agent, migrated in place

emrg/gui/main.js

@@ -968,15 +968,15 @@ vision = false
 
     // ── Task/template CRUD IPC (rant 2026-08-12T18:23:15 P3) ──────────
     ipcMain.handle("emrg:taskCreate", async (_e, payload) => {
-      const { name, type, project, interval, enabled, repo, description, sandbox, schedule } = payload || {};
+      const { name, type, agent, interval, enabled, repo, description, sandbox, schedule } = payload || {};
       if (typeof name !== "string" || !name.trim()) throw new Error("invalid task name");
       if (typeof type !== "string" || !type.trim()) throw new Error("invalid task type");
-      if (typeof project !== "string" || !project.trim()) throw new Error("invalid project");
+      if (typeof agent !== "string" || !agent.trim()) throw new Error("invalid agent");
       // `schedule` 必须原样转发：这张固定字段表此前把它丢掉，于是 GUI 永远只能建
       // 单一频次的任务，daemon 收了也白收（issue #2490）。只有 payload 真的带了这个
       // 键才发，缺席＝不声明，与 task_update 的「缺省不动该字段」一致。
       const fields = {
-        name: name.trim(), task_type: type.trim(), project: project.trim(),
+        name: name.trim(), task_type: type.trim(), agent: agent.trim(),
         interval, enabled, repo, description, sandbox,
       };
       if ("schedule" in (payload || {})) fields.schedule = schedule;

emrg/gui/renderer/src/components/TaskFormDialog.test.tsx

@@ -37,7 +37,7 @@ function taskWith(overrides: Partial<TaskRec> = {}): TaskRec {
     type: "evolution",
     interval: 10800,
     enabled: true,
-    config: { project: "emrg" },
+    config: { agent: "emrg" },
     ...overrides,
   };
 }
@@ -91,7 +91,7 @@ describe("TaskFormDialog", () => {
   });
 
   it("编辑：预填任务字段，name 只读不可改名", () => {
-    const task = taskWith({ type: "journal", interval: 3600, enabled: false, config: { project: "sci" } });
+    const task = taskWith({ type: "journal", interval: 3600, enabled: false, config: { agent: "sci" } });
     (task as TaskRec & { sandbox?: string }).sandbox = "read-only";
     render(
       wrapper(
@@ -161,7 +161,7 @@ describe("TaskFormDialog", () => {
     expect(payload).toEqual({
       name: "t1",
       type: "journal",
-      project: "sci",
+      agent: "sci",
       interval: 300,
       enabled: false,
       repo: undefined,

emrg/gui/renderer/src/components/TaskFormDialog.tsx

@@ -19,7 +19,7 @@ import { Dialog } from "./Dialog";
 export interface TaskFormPayload {
   name: string;
   type: string;
-  project: string;
+  agent: string;
   interval: number;
   enabled: boolean;
   repo?: string;
@@ -114,8 +114,8 @@ export function TaskFormDialog({
     const curType =
       (task?.type && types.includes(task.type) ? task.type : types[0]) || "evolution";
     if (typeRef.current) typeRef.current.value = curType;
-    const curProj = projects.some((p) => (p.name || p.path || "") === task?.config?.project)
-      ? (task?.config?.project as string)
+    const curProj = projects.some((p) => (p.name || p.path || "") === task?.config?.agent)
+      ? (task?.config?.agent as string)
       : (projects[0]?.name || projects[0]?.path || "");
     if (projectRef.current) projectRef.current.value = curProj;
     if (intervalRef.current) intervalRef.current.value = String(task?.interval ?? 1800);
@@ -176,7 +176,7 @@ export function TaskFormDialog({
     const payload: TaskFormPayload = {
       name,
       type: typeRef.current.value,
-      project: projectRef.current.value,
+      agent: projectRef.current.value,
       interval,
       enabled: enabledRef.current?.checked ?? true,
       repo: repoRef.current?.value.trim() || undefined,

emrg/gui/renderer/src/components/WorkspaceView.test.tsx

@@ -82,7 +82,7 @@ describe("WorkspaceView", () => {
     expect(screen.getByTestId("projects-empty")).toBeInTheDocument();
 
     const projects = [project("emrg"), project("aitokenpool", "/scm/aitokenpool")];
-    const tasks = [{ name: "emrg-task", config: { project: "emrg" } }] as TaskRec[];
+    const tasks = [{ name: "emrg-task", config: { agent: "emrg" } }] as TaskRec[];
     const { rerender } = setup({ projects, tasks });
     expect(screen.getAllByTestId("project-row")).toHaveLength(2);
     // emrg 有演化任务 → ⚡ 徽标；aitokenpool 无

emrg/gui/renderer/src/lib/workspaceView.test.ts

@@ -117,10 +117,10 @@ describe("preprocessRantMarkdown", () => {
 });
 
 describe("buildEvolveProjects", () => {
-  it("收集 config.project 集合（含 evolution 任务）", () => {
+  it("收集 config.agent 集合（含 evolution 任务）", () => {
     const tasks: TaskRec[] = [
-      { name: "a", config: { project: "emrg" } },
-      { name: "b", config: { project: "aitokenpool" } },
+      { name: "a", config: { agent: "emrg" } },
+      { name: "b", config: { agent: "aitokenpool" } },
       { name: "c", config: {} },
     ];
     const set = buildEvolveProjects(tasks);
@@ -176,7 +176,7 @@ describe("taskStatusBadge", () => {
 
 describe("taskHint", () => {
   it("项目 / 间隔 / 停用 join ·", () => {
-    const h = taskHint({ config: { project: "emrg" }, interval: 1800 }, t);
+    const h = taskHint({ config: { agent: "emrg" }, interval: 1800 }, t);
     expect(h).toBe('emrg · app.taskInterval={"n":1800}');
     expect(taskHint({ enabled: false }, t)).toBe("app.taskDisabled");
   });

emrg/gui/renderer/src/lib/workspaceView.ts

@@ -44,7 +44,7 @@ export interface TaskRec {
   interval?: number | null;
   last_run_at?: string | null;
   saturation?: { heartbeat_active?: boolean; heartbeat_interval?: number } | null;
-  config?: { project?: string; repo?: string; sandbox?: string } | null;
+  config?: { agent?: string; repo?: string; sandbox?: string } | null;
   /** 分时段频次窗口；缺省＝全天单一 `interval`（`list_tasks()` 逐字回报） */
   schedule?: ScheduleWindow[] | null;
   /** 任务面板「打开会话」所需字段（daemon TaskHandler.status() 返回，vanilla #924 迁移时丢失） */
@@ -142,11 +142,11 @@ export function preprocessRantMarkdown(text?: string | null): string {
 
 /* ── 项目面板工具 ────────────────────────────────────────── */
 
-/** auto_evolve 徽标集合：tasks.yml 中 config.project 与项目名相同（含 evolution 类型） */
+/** auto_evolve 徽标集合：tasks.yml 中 config.agent 与项目名相同（含 evolution 类型） */
 export function buildEvolveProjects(tasks: TaskRec[] | null | undefined): Set<string> {
   const set = new Set<string>();
   for (const t of tasks || []) {
-    const proj = t?.config?.project;
+    const proj = t?.config?.agent;
     if (proj) set.add(proj);
   }
   return set;
@@ -202,7 +202,7 @@ export function taskStatusBadge(t: TaskRec, tFn?: TFunc, now: number = Date.now(
 export function taskHint(t: TaskRec, tFn?: TFunc): string {
   const tr = tFn || ((k: string) => k);
   const parts: string[] = [];
-  if (t.config?.project) parts.push(t.config.project);
+  if (t.config?.agent) parts.push(t.config.agent);
   if (t.interval != null) parts.push(tr("app.taskInterval", { n: t.interval }));
   if (t.enabled === false) parts.push(tr("app.taskDisabled"));
   return parts.join(" · ");

emrg/server/__main__.py

@@ -9,6 +9,7 @@
 from emrg.logfiles import migrate_legacy_logs
 from emrg.server.daemon import DaemonExit, _non_regular_kind, run_server
 from emrg.server.logcontext import session_label
+from emrg.server.scheduler import migrate_task_config_keys
 from emrg.tool_path import ensure_tool_dirs
 
 
@@ -202,6 +203,13 @@ def main() -> None:
         # fails without touching the host's register at all, and it sits in no path
         # derivation — asking where the register lives must never move it.
         migrate_legacy_registry()
+        # The task table's agent key moved from `project` to `agent` (issue #2531),
+        # and it moves here for the same reason: this is a process that owns the
+        # file, it runs before the scheduler is constructed below, and no test path
+        # reaches it — so a suite run cannot rewrite a host's live table under a
+        # daemon still reading the old key. Asking a record for its agent never
+        # writes; only this call does.
+        migrate_task_config_keys()
         result = asyncio.run(run_server(config.llm))
     except KeyboardInterrupt:
         # SIGINT delivered at the event-loop poll point escapes the main

emrg/server/competition_prompt.md

@@ -17,7 +17,7 @@ You are EMRG's competition participation module. **Every cycle you MUST fully ex
 5. **Think before acting.** Before an irreversible or quota-consuming action — a submission, a registration, starting a long run — state in the reply **what will be done, on what evidence, the expected result, and the cost of failure**. If the evidence is not there yet, go and get it rather than doing it to find out.
 
 ### Current State
-- Task project: **{{ task.project }}** (from tasks.yml)
+- Task agent: **{{ task.agent }}** (from tasks.yml)
 - Local source: `{{ source_dir }}`
 - **Current time: `{{ timestamp }}`（{{ current_time_human }}）** — the time anchor for judging deadlines
 
@@ -58,7 +58,7 @@ Cross-round continuity comes from what the daemon already gives you: it loads th
 
 Rants are how the host sends instructions directly to a task.
 
-- A rant matches **only if its `project` field is exactly `{{ task.project }}`**. Rants that do not match are ignored completely.
+- A rant matches **only if its `project` field is exactly `{{ task.agent }}`**. Rants that do not match are ignored completely.
 - Unmatched but seemingly relevant rants must be **counted and reported to the host** — never silently dropped.
 - A matching rant is an instruction for this round: it takes priority over the round's planned work.
 

emrg/server/daemon.py

@@ -2937,7 +2937,7 @@ async def _process_message(
                 if self._scheduler:
                     ok, res = self._scheduler.task_create(
                         name=name, task_type="evolution",
-                        project=name, interval=600,
+                        agent=name, interval=600,
                     )
                     if ok:
                         self._scheduler.reconcile()
@@ -3004,7 +3004,7 @@ async def _process_message(
             ok, res = self._scheduler.task_create(
                 name=msg.get("name", "").strip(),
                 task_type=msg.get("task_type", "").strip(),
-                project=msg.get("project", "").strip(),
+                agent=msg.get("agent", "").strip(),
                 interval=msg.get("interval"),
                 enabled=msg.get("enabled", True),
                 repo=msg.get("repo"),
@@ -3026,7 +3026,7 @@ async def _process_message(
             if not self._scheduler:
                 await self._send(ws, {"type": "task_result", "error": "scheduler not running"})
                 return
-            fields = {k: msg[k] for k in ("task_type", "project", "interval", "enabled", "repo", "description", "sandbox", "schedule") if k in msg}
+            fields = {k: msg[k] for k in ("task_type", "agent", "interval", "enabled", "repo", "description", "sandbox", "schedule") if k in msg}
             if "task_type" in fields:
                 fields["type"] = fields.pop("task_type")
             ok, res = self._scheduler.task_update(msg.get("name", "").strip(), **fields)

emrg/server/evolution_prompt.md

@@ -221,7 +221,7 @@ Two ways this source goes quiet, and both are legitimate: an installation whose
 - **Issue**: something needing a reply or triage, or a resolved one to close — the close itself is §3.1. What finishes an issue, and how the link reading is read, is R5.
 - **Rant**: take each one through three questions.
   - **Has it been handled?** `git log --oneline -20` for a commit naming its timestamp or keywords. Naming proves only that it was **touched**, never that it is finished — check for unmet acceptance items, unmerged branches, and a stage that landed without finishing the work.
-  - **Does it match this task?** A rant's `project` matches if it is `{{ task.project }}` or `{{ owner }}/{{ repo }}`; a rant that names neither is not this instance's to act on — **ignore rants without a `project` field entirely**.
+  - **Does it match this task?** A rant's `project` matches if it is `{{ task.agent }}` or `{{ owner }}/{{ repo }}`; a rant that names neither is not this instance's to act on — **ignore rants without a `project` field entirely**.
   - **Is it work?** A rant this cycle decides against is closed with the reason in its progress; a rant taken on becomes an issue before any code is written (R5). The lifecycle is R6. **Before any reason is written down as the host's ruling, R7.**
 
 #### 2.3 Decide this cycle's work

emrg/server/feishu_prompt.md

@@ -4,7 +4,7 @@ You are EMRG handling the host's Feishu messages: read what arrived, judge each
 what needs no permission, ask where it does.
 
 ### Current State
-- Task project: **{{ task.project }}** (from tasks.yml)
+- Task agent: **{{ task.agent }}** (from tasks.yml)
 - Local source: `{{ source_dir }}`
 - **Current time: `{{ timestamp }}`（{{ current_time_human }}）** — the anchor for "since the last round" and for every dated record
 
@@ -47,7 +47,7 @@ in the usual install locations before calling it blocked.
   task instructions.
 - Take the messages that arrived after the last one you handled. Remember that boundary
   yourself; there is no state file to keep and none to trust.
-- Rants whose `project` is `{{ task.project }}` are the host's instructions for this round
+- Rants whose `project` is `{{ task.agent }}` are the host's instructions for this round
   and outrank the plan; those that do not match are counted, not acted on.
 - A CLI that cannot run is a `blocked` item reported to the host, and so is an account that
   really cannot authenticate — but `missing identity` is a symptom, not a cause. The CLI

emrg/server/ios_app_prompt.md

@@ -12,7 +12,7 @@ You are EMRG as the **product owner** of iOS applications: you choose what to bu
 **Every round runs §0 Preparation first (§0.1–§0.7), whichever phase it then enters.**
 
 ### Current State
-- Task project: **{{ task.project }}** (from tasks.yml)
+- Task agent: **{{ task.agent }}** (from tasks.yml)
 - Local source: `{{ source_dir }}`
 - **Current time: `{{ timestamp }}`（{{ current_time_human }}）** — the time anchor for judging review turnaround and every dated record
 
@@ -60,7 +60,7 @@ Everything needed to build, sign and upload an iOS app is installed. **Adding a
 
 Rants are how the host sends instructions directly to a task.
 
-- A rant matches **only if its `project` field is exactly `{{ task.project }}`**. Rants that do not match are ignored completely.
+- A rant matches **only if its `project` field is exactly `{{ task.agent }}`**. Rants that do not match are ignored completely.
 - Unmatched but seemingly relevant rants must be **counted and reported to the host**, never silently dropped.
 - A matching rant is an instruction for this round and takes priority over the round's planned work.
 

emrg/server/journal_prompt.md

@@ -127,9 +127,9 @@ cat ~/.emrg/rants.jsonl 2>/dev/null || echo "[no rants.jsonl — skip rant scan]
 ```
 
 Filter rules (same as open-source tasks):
-- Match rant's `project` against exactly `{{ task.project }}` — equal counts, anything else does not
+- Match rant's `project` against exactly `{{ task.agent }}` — equal counts, anything else does not
 - Ignore rants without a `project` field; only `pending` / `in_progress` count
-- **Irrelevant-rant exclusion (HARD)**: any rant that does NOT match `project` exactly, or whose content is not about THIS journal, must be **ignored entirely** — do NOT read it into your thinking, do NOT adopt it as a research direction, do NOT cite it. **An irrelevant rant must never become a topic anchor.** Only a rant with `project` = `{{ task.project }}` AND content actually directed at this journal enters the candidate pool / Phase Ops.
+- **Irrelevant-rant exclusion (HARD)**: any rant that does NOT match `project` exactly, or whose content is not about THIS journal, must be **ignored entirely** — do NOT read it into your thinking, do NOT adopt it as a research direction, do NOT cite it. **An irrelevant rant must never become a topic anchor.** Only a rant with `project` = `{{ task.agent }}` AND content actually directed at this journal enters the candidate pool / Phase Ops.
 - **Dedup check** before treating any rant as actionable: search the journal repo commit log for the rant's timestamp/keywords — a commit referencing the rant only counts as handled when ALL acceptance items are satisfied AND related PRs merged
 - **Rant status machine**: `pending → in_progress → completed` (never jump pending → completed). When starting work on a rant: set `in_progress` + progress note. When all its PRs merged + self-verification passes: set `completed` + ISO timestamp. Host feedback that a fix is insufficient → revert to `in_progress` with reason.
 - Rant-driven journal ops (e.g. "adjust CfP", "revise review policy") are processed in Phase Ops with highest priority.
@@ -370,7 +370,7 @@ Nothing pending?                                      → Phase Research (new di
      - Read the 2–3 most relevant recent abstracts (last 6 months), record their limitations → this forms your gap. If browser-harness is available, prefer it for reading full texts.
    - **The editor's CfP / journal themes**: README, CfP issue, editorial comments — topics the journal explicitly wants.
    - **Your own papers' future work**: Discussion / Future-Work paragraphs of your published or in-review manuscripts. **Own-work relevance is NOT by itself a topic reason** — it is only one candidate among many.
-   - **Host rants with `project` = {{ task.project }}**: directional feedback, not just bug reports. (Per §0.5, irrelevant rants are excluded entirely.)
+   - **Host rants with `project` = {{ task.agent }}**: directional feedback, not just bug reports. (Per §0.5, irrelevant rants are excluded entirely.)
    - **Uncovered gaps in the journal**: open issues / published papers that raise questions nobody has answered yet.
    - **Data-source debias**: having data for a particular project on hand is NOT a topic reason. Prefer directions backed by external data sources (public datasets, open-source repositories, simulations) or where data can be obtained/created during the research.
 2. **Research-hotspot identification** (using the current-time anchor; run in parallel with the candidate pool):
@@ -544,7 +544,7 @@ cd {{ source_dir }} && gh issue list -R {{ owner }}/{{ repo }} --label in-review
 
 1. **Never touch host dirt**: never stash/reset/clean host work — the daemon converges the tree itself before the prompt, so a tree still dirty here is a convergence that failed (and the round is then read-only); `papers/*/research/` being present is normal
 2. **git add exact path**: submission/revision commits use `git add papers/issue-<N>/` — never `-A` / `.`
-3. **Rant handling**: follow 0.5 — project must equal `{{ task.project }}`; pending → in_progress → completed
+3. **Rant handling**: follow 0.5 — project must equal `{{ task.agent }}`; pending → in_progress → completed
 4. **Language policy**: external journal-facing text (issues/PRs/reviews/decisions) in English; internal records (session history, memory entries, the closing summary) in the author's language
 5. **One thing at a time**: advance exactly one phase per cycle; don't aim for completeness, just for progress
 6. **Never claim/review your own submission** (author)