feat: retry main agent runs

2026-03-29 22:59:15 +08:00
parent 568e8091c1
commit 32c28fb7d6
4 changed files with 172 additions and 0 deletions
--- a/collector-service/app/oneliner_features.py
+++ b/collector-service/app/oneliner_features.py
@@ -182,6 +182,10 @@ class AgentRunCancelRequest(BaseModel):
    reason: str = ""


+class AgentRunRetryRequest(BaseModel):
+    reason: str = ""
+
+
 INTENT_ACTIONS: dict[str, list[dict[str, Any]]] = {
    "create_project": [{"key": "goto-intake", "label": "去我的项目", "kind": "navigate"}],
    "create_assistant": [{"key": "open-create-assistant", "label": "创建 Agent", "kind": "ui_action"}],
@@ -4811,6 +4815,92 @@ def register_oneliner_routes(app: Any, legacy: Any) -> None:
        assert row is not None
        return _agent_run_payload(row)

+    @app.post("/v2/oneliner/runs/{run_id}/retry")
+    def retry_oneliner_run(
+        run_id: str,
+        request: AgentRunRetryRequest,
+        account: dict[str, Any] = Depends(legacy.require_approved),
+    ) -> dict[str, Any]:
+        row = _load_owned_agent_run(run_id, account)
+        current_status = str(row.get("run_status") or "")
+        if current_status not in {"cancelled", "blocked", "failed", "done"}:
+            raise HTTPException(status_code=409, detail="Run can not be retried yet")
+
+        project_id = str(row.get("project_id") or "").strip()
+        platform = str(row.get("platform") or "").strip()
+        platform_scope = _normalize_platform_scope(str(row.get("platform_scope") or "single_platform"))
+        governance = _effective_policy_payload(
+            subject_account=account,
+            subject_project_id=project_id,
+            platform=platform,
+        )
+        plan = _parse_json(row.get("plan_json"), {})
+        session_id = str(row.get("session_id") or "").strip()
+        _touch_session_for_run(session_id, platform=platform, intent_key=str(row.get("intent_key") or "custom").strip() or "custom")
+
+        next_run_id = make_id("oline_run")
+        timestamp = now()
+        active_admin_override_notice = governance.get("active_admin_override_notice") or {}
+        legacy.db.execute(
+            """
+            INSERT INTO agent_runs (
+                id, user_id, project_id, session_id, source_screen, source_action_key, title, summary,
+                intent_key, platform, platform_scope, delivery_mode, run_status, scheduling_mode,
+                active_executor_key, plan_json, governance_json, result_json, status_summary,
+                needs_user_input, blocked_reason, active_admin_override_notice_json,
+                created_at, updated_at, started_at, finished_at
+            ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 'needs_confirmation', ?, 'main_agent', ?, ?, '{}', ?, 1, '', ?, ?, ?, '', '')
+            """,
+            (
+                next_run_id,
+                account["id"],
+                project_id,
+                session_id,
+                str(row.get("source_screen") or "").strip(),
+                str(row.get("source_action_key") or "").strip(),
+                str(row.get("title") or plan.get("goal") or "主 Agent 任务").strip() or "主 Agent 任务",
+                str(request.reason or row.get("summary") or "").strip() or str(row.get("summary") or "").strip(),
+                str(row.get("intent_key") or "custom").strip() or "custom",
+                platform,
+                platform_scope,
+                _normalize_delivery_mode(str(row.get("delivery_mode") or "hybrid")),
+                _normalize_scheduling_mode(str(row.get("scheduling_mode") or "queued")),
+                _dump(plan),
+                _dump(governance),
+                "等待你确认执行计划",
+                _dump(active_admin_override_notice),
+                timestamp,
+                timestamp,
+            ),
+        )
+        _log_agent_run_event(
+            next_run_id,
+            event_type="run.created",
+            summary=f"已创建待确认任务：{str(row.get('title') or plan.get('goal') or '主 Agent 任务').strip() or '主 Agent 任务'}",
+            details={
+                "run_status": "needs_confirmation",
+                "source_screen": str(row.get("source_screen") or "").strip(),
+                "source_action_key": str(row.get("source_action_key") or "").strip(),
+                "platform": platform,
+                "platform_scope": platform_scope,
+            },
+        )
+        _log_agent_run_event(
+            next_run_id,
+            event_type="run.retried",
+            summary=str(request.reason or "已基于上一轮主 Agent 任务重新生成待确认执行卡").strip(),
+            details={"retry_from_run_id": run_id, "from_status": current_status},
+        )
+        _log_agent_run_event(
+            run_id,
+            event_type="run.retried",
+            summary=str(request.reason or "用户要求重新发起当前主 Agent 任务").strip(),
+            details={"new_run_id": next_run_id, "from_status": current_status},
+        )
+        created = legacy.db.fetch_one("SELECT * FROM agent_runs WHERE id = ?", (next_run_id,))
+        assert created is not None
+        return _agent_run_payload(created)
+
    @app.get("/v2/oneliner/runs")
    def list_oneliner_runs(
        project_id: str | None = Query(default=None),
--- a/tests/test_main_agent_governance.py
+++ b/tests/test_main_agent_governance.py
@@ -299,6 +299,54 @@ class MainAgentGovernanceTests(unittest.TestCase):
        self.assertIn("run.progress", event_types)
        self.assertIn("run.done", event_types)

+    def test_cancelled_run_can_be_retried_as_a_new_pending_run(self) -> None:
+        create = self.client.post(
+            "/v2/oneliner/runs",
+            headers=self.ctx["member_headers"],
+            json={
+                "project_id": self.ctx["project_id"],
+                "source_screen": "production",
+                "source_action_key": "handoff-to-main-agent",
+                "title": "恢复失败任务",
+                "summary": "让主 Agent 重新接单",
+                "intent_key": "custom",
+                "platform": "douyin",
+                "platform_scope": "single_platform",
+                "plan_request": {
+                    "goal": "恢复失败任务",
+                    "steps": ["检查失败原因", "重新安排执行", "确认落点"],
+                },
+            },
+        )
+        self.assertEqual(create.status_code, 200, create.text)
+        original_run = create.json()
+        run_id = original_run["id"]
+
+        cancel = self.client.post(
+            f"/v2/oneliner/runs/{run_id}/cancel",
+            headers=self.ctx["member_headers"],
+            json={"reason": "user cancelled"},
+        )
+        self.assertEqual(cancel.status_code, 200, cancel.text)
+        self.assertEqual(cancel.json()["run_status"], "cancelled")
+
+        retry = self.client.post(
+            f"/v2/oneliner/runs/{run_id}/retry",
+            headers=self.ctx["member_headers"],
+            json={"reason": "retry from runtime"},
+        )
+        self.assertEqual(retry.status_code, 200, retry.text)
+        payload = retry.json()
+        self.assertNotEqual(payload["id"], run_id)
+        self.assertEqual(payload["run_status"], "needs_confirmation")
+        self.assertEqual(payload["title"], original_run["title"])
+        self.assertEqual(payload["project_id"], self.ctx["project_id"])
+        self.assertEqual(payload["plan"]["goal"], "恢复失败任务")
+        self.assertEqual(payload["recommended_preview_action"]["screen"], "production")
+        event_types = [item["event_type"] for item in payload["events"]]
+        self.assertIn("run.created", event_types)
+        self.assertIn("run.retried", event_types)
+
    def test_effective_policy_merges_system_user_global_and_platform_layers(self) -> None:
        system_response = self.client.put(
            "/v2/admin/oneliner/governance/system/main-agent",
--- a/web/storyforge-web-v4/assets/app.js
+++ b/web/storyforge-web-v4/assets/app.js
@@ -1029,6 +1029,7 @@ function renderOneLinerRunsHtml() {
    : currentRun.run_status === "failed"
    ? "orange"
    : "";
+  const canRetryCurrentRun = ["blocked", "failed", "cancelled"].includes(currentRun.run_status);
  return `
    <div class="task-item compact" style="margin-bottom:10px;">
      <h4>近期运行概况</h4>
@@ -1113,6 +1114,7 @@ function renderOneLinerRunsHtml() {
        ` : `
          <span class="tag ${statusTone}">${escapeHtml(currentRun.status_summary || "主 Agent 正在推进中")}</span>
        `}
+        ${canRetryCurrentRun ? `<span class="tag clickable-tag" data-action="retry-oneliner-run" data-run-id="${escapeHtml(currentRun.id)}">重新执行</span>` : ""}
        ${hasResultPayload ? `<span class="tag clickable-tag" data-action="open-oneliner-run-result" data-run-id="${escapeHtml(currentRun.id)}">查看结果</span>` : ""}
        ${recommendedAction?.action ? `<span class="tag clickable-tag" data-action="${escapeHtml(recommendedAction.action)}" ${resultLandingAttrs}>${escapeHtml(recommendedAction.label || "回到对应页面")}</span>` : ""}
      </div>
@@ -1848,6 +1850,18 @@ async function cancelOneLinerRun(runId, reason = "") {
  return payload;
 }

+async function retryOneLinerRun(runId, reason = "") {
+  const payload = await storyforgeFetch(`/v2/oneliner/runs/${encodeURIComponent(runId)}/retry`, {
+    method: "POST",
+    body: { reason }
+  });
+  await loadAgentControlSurfaces(getOneLinerProjectId());
+  appState.selectedOnelinerRunId = payload?.id || choosePreferredOneLinerRunId(appState.onelinerRuns, "");
+  appState.onelinerRunFilter = "focus";
+  rememberAction("主 Agent 已重新生成待确认卡", payload?.title || payload?.plan?.goal || "你可以先确认新的执行计划。", "green", payload);
+  return payload;
+}
+
 function renderOneLinerExecutionPayloadHtml(payload) {
  if (!payload || typeof payload !== "object") {
    return `<div class="task-item compact"><h4>没有返回执行结果</h4><p>当前执行器没有附带额外数据。</p></div>`;
@@ -10020,6 +10034,18 @@ document.addEventListener("click", async (event) => {
      }
      return;
    }
+    if (name === "retry-oneliner-run") {
+      try {
+        setBusy(true, "正在重新生成待确认卡...");
+        await retryOneLinerRun(action.dataset.runId || "", "user requested retry");
+        openOneLinerPanel();
+      } catch (error) {
+        presentActionFailure(error, "主 Agent 重开失败");
+      } finally {
+        setBusy(false, "");
+      }
+      return;
+    }
    if (name === "select-oneliner-run") {
      appState.selectedOnelinerRunId = action.dataset.runId || "";
      renderAll();
--- a/web/storyforge-web-v4/tests/workbench-pages.test.mjs
+++ b/web/storyforge-web-v4/tests/workbench-pages.test.mjs
@@ -329,6 +329,14 @@ test("opening a main agent run result keeps that run selected in the floating ru
  assert.match(resultAction, /appState\.onelinerRunFilter = currentRun\.run_status === "done" \? "done" : appState\.onelinerRunFilter/);
 });

+test("oneliner runtime exposes retry for retryable runs and wires the action handler", () => {
+  const runtime = extractBetween(APP, "function renderOneLinerRunsHtml()", "function renderOneLinerMessagesHtml()");
+  const actions = extractBetween(APP, "document.addEventListener(\"click\", async (event) => {", "document.addEventListener(\"submit\", async (event) => {");
+  assert.match(runtime, /retry-oneliner-run/);
+  assert.match(actions, /name === "retry-oneliner-run"/);
+  assert.match(actions, /await retryOneLinerRun\(action\.dataset\.runId \|\| "", "user requested retry"\)/);
+});
+
 test("oneliner panel auto-polls active runs while the floating panel stays open", () => {
  const render = extractBetween(APP, "function renderOneLinerUi()", "function openOneLinerPanel()");
  const open = extractBetween(APP, "function openOneLinerPanel()", "function closeOneLinerPanel()");