@sleep2agi/commhub-server 0.9.0-preview.36 → 0.9.0-preview.38

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sleep2agi/commhub-server",
-  "version": "0.9.0-preview.36",
+  "version": "0.9.0-preview.38",
   "description": "CommHub Server — AI Agent communication hub with MCP protocol, multi-network isolation, user auth, and MCP tools (17 collaboration-core + node/provider ops tools; authoritative list: docs-site/docs/api/mcp-tools.md).",
   "type": "module",
   "main": "src/index.ts",

package/src/create-node.ts

@@ -306,6 +306,7 @@ export function auditCreateNodeStrict(input: {
 export function auditCreateNode(input: {
   action:
     | "create_node_dispatched"
+    | "create_node_dispatch_attempted"
     | "create_node_rejected"
     | "create_node_succeeded"
     | "create_node_sweeper_revoked"

package/src/stop-delete-node.test.ts

@@ -36,6 +36,7 @@ interface Reply { ok?: boolean; error?: string; [k: string]: unknown; }
 function cleanup() {
   for (const n of [NET_A, NET_B]) {
     try { db.run("DELETE FROM node_stop_requests WHERE network_id = ?1", [n]); } catch {}
+    try { db.run("DELETE FROM node_create_requests WHERE network_id = ?1", [n]); } catch {}
     try { db.run("DELETE FROM audit_log WHERE network_id = ?1", [n]); } catch {}
     try { db.run("DELETE FROM inbox WHERE network_id = ?1", [n]); } catch {}
     // PR1.2a restart_node tests leave node_config_updates rows; clear
@@ -676,6 +677,60 @@ describe("list_my_children HANDLER (RFC-027 PR1.1)", () => {
   });
 });
 
+describe("create_node audit wording (#1362)", () => {
+  test("create_node records a dispatch attempt, not guaranteed SSE delivery", async () => {
+    setupAlphaNetwork();
+    const tools = buildHandlers(USER_A_ID);
+    const r = await call(tools.create_node, {
+      daemon_node_id: DAEMON_A_ID,
+      node_spec: {
+        name: "sd-new-child",
+        runtime: "claude-agent-sdk",
+        model: "x",
+      },
+      network_id: NET_A,
+    });
+
+    expect(r.ok).toBe(true);
+    expect(typeof r.request_id).toBe("string");
+    expect(readAudit("create_node_dispatch_attempted", NET_A).length).toBe(1);
+    expect(readAudit("create_node_dispatched", NET_A).length).toBe(0);
+  });
+});
+
+describe("list_my_pending_create_requests HANDLER (#1362 SSE reconnect compensation)", () => {
+  test("daemon-bound caller gets only its own pending create requests", async () => {
+    setupAlphaNetwork();
+    const otherDaemonId = "node_sd_daemon_other";
+    seedDaemon(NET_A, otherDaemonId, "sd-daemon-other", USER_A_ID, "tok_sd_daemon_other");
+    db.run(
+      `INSERT INTO node_create_requests
+         (request_id, daemon_node_id, child_name, network_id, runtime, model, flags_json, env_keys, status, child_token_id, created_at, created_by_token)
+       VALUES
+         ('cr_pending_mine', ?1, 'child-pending-mine', ?2, 'claude-agent-sdk', 'x', '{}', '[]', 'pending', NULL, ?3, 'tok_test'),
+         ('cr_delivered_mine', ?1, 'child-delivered-mine', ?2, 'claude-agent-sdk', 'x', '{}', '[]', 'delivered', NULL, ?4, 'tok_test'),
+         ('cr_pending_other', ?5, 'child-pending-other', ?2, 'claude-agent-sdk', 'x', '{}', '[]', 'pending', NULL, ?6, 'tok_test')`,
+      [DAEMON_A_ID, NET_A, Date.now(), Date.now() + 1, otherDaemonId, Date.now() + 2],
+    );
+
+    const daemonTools = buildHandlers(USER_A_ID, true, DAEMON_A_TOK, NET_A);
+    const r = await call(daemonTools.list_my_pending_create_requests, {});
+
+    expect(r.ok).toBe(true);
+    expect(r.count).toBe(1);
+    expect(r.requests).toEqual([
+      expect.objectContaining({ request_id: "cr_pending_mine", child_name: "child-pending-mine" }),
+    ]);
+  });
+
+  test("non-daemon caller (utok) refused", async () => {
+    setupAlphaNetwork();
+    const userTools = buildHandlers(USER_A_ID);
+    const r = await call(userTools.list_my_pending_create_requests, {});
+    expect(r.ok).toBe(false);
+  });
+});
+
 // ─── PR1.2a — restart_node resets lifecycle_state ──────────────────
 //
 // 通信龙 #346 ack latent: without this, a stopped node restart-ed via
@@ -718,3 +773,60 @@ describe("RFC-027 PR1.2a — restart_node resets lifecycle_state to 'active'", (
     expect(readNode(CHILD_A_ID)?.lifecycle_state).toBe("active");
   });
 });
+
+// ── #1286 stale-deleting 重派出口 ──────────────────────────────────
+//
+// 修复前 daemon 丢 ack 会让行永远停在 deleting：重发被挡、force 也一样。
+// 2026-08-28 实测两行卡死（acc-…-3373 / probe-tl2），API 层面无解。
+// 出口三条件缺一不放行：force=true + 最近 delete 请求非终态 + 晾超 5 分钟。
+describe("delete_node — stale-deleting redispatch exit (#1286)", () => {
+  function seedDeletingChild() {
+    setupAlphaNetwork();
+    db.run(`UPDATE nodes SET lifecycle_state = 'deleting' WHERE node_id = ?1`, [CHILD_A_ID]);
+  }
+  function seedStopRequest(status: string, ageMs: number) {
+    db.run(
+      `INSERT INTO node_stop_requests (request_id, network_id, daemon_node_id, child_node_id,
+         child_alias, action, delete_config, created_by_token, status, created_at)
+       VALUES (?1, ?2, ?3, ?4, ?5, 'delete', 1, 'tok_test', ?6, ?7)`,
+      [`sr_stale_${Date.now()}_${Math.floor(ageMs)}`, NET_A, DAEMON_A_ID, CHILD_A_ID,
+       CHILD_A_ALIAS, status, Date.now() - ageMs],
+    );
+  }
+  const base = { child_node_id: CHILD_A_ID, daemon_node_id: DAEMON_A_ID, network_id: NET_A, confirm_alias: CHILD_A_ALIAS };
+
+  test("无 force：仍拒，且带 hint 指出口（不是让人反复撞墙）", async () => {
+    seedDeletingChild(); seedStopRequest("pending", 10 * 60_000);
+    const r = await call(buildHandlers(USER_A_ID).delete_node, base);
+    expect(r.ok).toBe(false);
+    expect(r.error).toBe("node_already_deleting");
+    expect(typeof r.hint).toBe("string");
+  });
+
+  test("🔴 force 但请求还新鲜（1 分钟）：仍拒 —— force 不能变成竞态放大器", async () => {
+    seedDeletingChild(); seedStopRequest("pending", 60_000);
+    const r = await call(buildHandlers(USER_A_ID).delete_node, { ...base, force: true });
+    expect(r.ok).toBe(false);
+    expect(r.error).toBe("node_already_deleting");
+  });
+
+  test("🔴 force 但上一条已终态（stopped）：仍拒 —— daemon 刚 ack 过，行马上会转移", async () => {
+    seedDeletingChild(); seedStopRequest("stopped", 10 * 60_000);
+    const r = await call(buildHandlers(USER_A_ID).delete_node, { ...base, force: true });
+    expect(r.ok).toBe(false);
+    expect(r.error).toBe("node_already_deleting");
+  });
+
+  test("force + pending 晾超 5 分钟：放行，生成新 request_id", async () => {
+    seedDeletingChild(); seedStopRequest("pending", 10 * 60_000);
+    const r = await call(buildHandlers(USER_A_ID).delete_node, { ...base, force: true });
+    expect(r.ok).toBe(true);
+    expect((r.request_id as string).startsWith("sr_")).toBe(true);
+  });
+
+  test("force + dispatched 晾超 5 分钟：同样放行", async () => {
+    seedDeletingChild(); seedStopRequest("dispatched", 10 * 60_000);
+    const r = await call(buildHandlers(USER_A_ID).delete_node, { ...base, force: true });
+    expect(r.ok).toBe(true);
+  });
+});

package/src/tools.ts

@@ -3215,9 +3215,10 @@ export function registerTools(server: McpServer, clientIP?: string, enforceNetwo
       // SSE handler resolves the rest via MCP call.
       pushEvent(daemon.alias, { type: "create_node", request_id: requestId }, networkIdForChild);
 
-      // §4.5 audit — dispatch succeeded
+      // §4.5 audit — request queued and doorbell attempted. pushEvent is
+      // fire-and-forget, so actual SSE delivery is intentionally not claimed.
       auditCreateNode({
-        action: "create_node_dispatched",
+        action: "create_node_dispatch_attempted",
         user_id: enforceUserId,
         network_id: networkIdForChild,
         target_id: requestId,
@@ -3237,6 +3238,31 @@ export function registerTools(server: McpServer, clientIP?: string, enforceNetwo
     },
   );
 
+  // #1362 — daemon-side compensation for missed SSE doorbells. pushEvent is
+  // intentionally best-effort; after a daemon connects or reconnects, it pulls
+  // any still-pending requests bound to its own daemon token.
+  server.tool(
+    "list_my_pending_create_requests",
+    "Daemon lists pending create-node requests bound to itself for SSE reconnect compensation. RFC-026/#1362.",
+    {},
+    async () => {
+      const callerDaemon = resolveCallerDaemonTokenBound();
+      if (!callerDaemon.ok) {
+        return { content: [{ type: "text" as const, text: JSON.stringify({ ok: false, error: callerDaemon.error }) }] };
+      }
+      const rows = db.all<{ request_id: string; child_name: string; created_at: number }>(
+        `SELECT request_id, child_name, created_at
+           FROM node_create_requests
+          WHERE daemon_node_id = ?1
+            AND network_id = ?2
+            AND status = 'pending'
+          ORDER BY created_at ASC`,
+        callerDaemon.daemonNodeId, callerDaemon.networkId,
+      );
+      return { content: [{ type: "text" as const, text: JSON.stringify({ ok: true, count: rows.length, requests: rows }) }] };
+    },
+  );
+
   // §2.5 step 3 — daemon-facing pull. Returns full spec + env_blob +
   // child_ntok in one shot. Map evicted on take (one-shot consume).
   //
@@ -3561,7 +3587,39 @@ export function registerTools(server: McpServer, clientIP?: string, enforceNetwo
       return { ok: false, error: state === "stopping" ? "node_already_stopping" : "node_not_active", current_state: state };
     }
     if (args.action === "delete" && state === "deleting") {
-      return { ok: false, error: "node_already_deleting", current_state: state };
+      // #1286 —— stale-deleting 重派出口。
+      //
+      // 在 .40 埋点修复之前,daemon 丢失 ack 会让行**永远**停在 deleting:
+      // 重发 delete 被这里挡住,force=true 也一样 —— 状态机没有出口,
+      // 2026-08-28 实测两行卡死(acc-…-3373 / probe-tl2),API 层面无解。
+      //
+      // 出口的三个条件,缺一不放行:
+      //   ① 调用方显式 force=true(不改默认行为 —— 正常并发的重复 delete 仍被拒)
+      //   ② 该 child 最近一条 delete 请求不是终态(pending/dispatched)
+      //      —— 终态说明 daemon 刚 ack 过,行马上会转移,不该重派
+      //   ③ 那条请求已经晾了超过 STALE_DELETING_MS —— 🔴 不是"刚派发就能 force":
+      //      给正常的 doorbell→ack 往返留时间,否则 force 变成竞态放大器。
+      const STALE_DELETING_MS = 5 * 60_000;
+      const last = db.get<{ request_id: string; status: string; created_at: number }>(
+        `SELECT request_id, status, created_at FROM node_stop_requests
+          WHERE child_node_id = ?1 AND action = 'delete'
+          ORDER BY created_at DESC LIMIT 1`, node.node_id,
+      );
+      const lastIsStale = !!last
+        && (last.status === "pending" || last.status === "dispatched")
+        && (Date.now() - last.created_at) > STALE_DELETING_MS;
+      if (!(args.force === true && (lastIsStale || !last))) {
+        return {
+          ok: false, error: "node_already_deleting", current_state: state,
+          // 🔴 告诉调用方出口在哪,而不是让他反复撞同一面墙。
+          ...(last ? { last_request_id: last.request_id, last_request_status: last.status } : {}),
+          hint: last && (last.status === "pending" || last.status === "dispatched")
+            ? `再等一下;若超过 ${STALE_DELETING_MS / 60000} 分钟仍未收敛,带 force=true 重试可重新派发`
+            : "上一条请求已终态,行应即将转移;稍后重查",
+        };
+      }
+      // 放行:走正常派发路径,生成新 request_id。旧请求行保留(审计线索),
+      // daemon 侧对未知 request 的 get_stop_request 会拿到 request_not_found,无副作用。
     }
     if (args.action === "delete" && state === "stopping") {
       // Can't start delete while a stop is mid-flight — race-prone.