Cloud agent runs fail immediately with stream_unavailable / “Run stream is no longer available”; agent often deleted (404) after first failed bootstrap

Ilan_Levy · September 12, 2026, 6:17am

Where does the bug appear (feature/product)?

Cursor SDK

Describe the Bug

Using @cursor/sdk (^1.0.28), Agent.create() succeeds and agent.send(prompt) returns a run in running state, but the run transitions to error within a few seconds with no assistant output. run.wait() often returns error with error.code: “stream_unavailable” and error.message: “Run stream is no longer available”. durationMs is sometimes null. Shortly after, the same agentId is not found (agent appears removed).

This reproduces for repo-only cloud config and for named cloud environments; it is not caused by sending env.name and repos together (we only send one mode).

Environment
SDK: @cursor/sdk ^1.0.28 (Node, ESM, tsx)
Model: { id: “composer-2.5” }
Client: Internal workflow orchestrator (Agent.create → agent.send → run.stream() + run.wait())
OS: macOS (orchestrator local); agents run on Cursor cloud
Auth: Team API key (CURSOR_API_KEY)

Steps to Reproduce

Create an isolated cloud agent:
await using agent = await Agent.create({
apiKey,
model: { id: “composer-2.5” },
name: “workflow-AE-50228”,
cloud: {
skipReviewerRequest: true,
envVars: { /* team personal keys, no secrets in ticket / },
repos: [{
url: “https://gitlab.com/agora.re/app/agora-api-monorepo”,
startingRef: “development”,
}],
// OR named env only (no repos):
// env: { type: “cloud”, name: “ilan” },
},
});
Send first-turn prompt (~4.5k chars — ticket bootstrap / skills; not unusually large).
const run = await agent.send(prompt);
// run.status === “running”
Consume stream and wait:
for await (const event of run.stream()) { / assistant text */ }
const result = await run.wait();
// result.status === “error”
Optionally resume/fetch the same agent.agentId — often 404 / not found.
Variant B — named environment only

cloud: {
env: { type: “cloud”, name: “ilan” }, // multi-repo env configured in Cursor dashboard
repos: undefined,
}
Same failure pattern.

Expected Behavior

Run completes (finished) or fails with a actionable API error (auth, repo access, env misconfiguration, prompt rejection, etc.).
Stream stays available until terminal run state, or wait() returns a stable error payload.
Agent remains resumable on Agent.resume(agentId) after a failed run unless explicitly deleted.
Actual behavior
Run goes running → error in ~1–10s.
Zero streamed assistant tokens; result.result empty.
run.wait() error (when present):
{
“message”: “Run stream is no longer available”,
“code”: “stream_unavailable”
}
durationMs sometimes null.
After failure, health check / Agent.resume often indicates agent missing (probe=missing in our logs).
Evidence (recent repro — repo-only)
Field Value
Ticket / metadata
AE-50228
agentId
bc-dbaaba57-e59b-4dcb-bc75-a61ef2aadcea
runId
run-dd4e3bb8-de6a-4e32-94ed-6a621f422cb0
Cloud payload
Repo-only: agora-api-monorepo @ development, no env.name
Prompt size
4504 chars (first turn)
Diagnostics
resultChars: 0, streamedChars: 0, errorCode: “stream_unavailable”
Post-failure
Agent not found
Other run IDs (same symptom):
run-e1a2c4eb-99c4-4261-9814-4ec3ff13478b (named env ilan),
run-0c7aa663-4b6d-40e8-a284-f5418867a7d0,
run-f6f03d79-da07-4afa-bc85-d0610f5718ee.

Not observed: [validation_error] env.name cannot be combined with repos on send (we enforce env XOR repos client-side).

What we ruled out on the client
Local orchestrator crash (stderr clean; send succeeds).
Accidental env + repos in create payload (logged SDK shape is correct).
Missing CURSOR_API_KEY / personal env injection (create succeeds; envVarKeys present).
Questions for SDK / platform
Is stream_unavailable on run.wait().error intended to mean the run failed on the server, or only local stream teardown? Should callers treat it as retryable?
Why is the agent deleted or 404 after a failed first run?
Is there a supported way to fetch server-side run failure reason when the stream closes early (e.g. getRun, webhook, or expanded RunResult)?
Any known limits on first-turn prompt size or bootstrap for repo-cloned cloud VMs that would kill the run without a clear error?
Minimal repro sketch
import { Agent } from “@cursor/sdk”;
const apiKey = process.env.CURSOR_API_KEY!;
await using agent = await Agent.create({
apiKey,
model: { id: “composer-2.5” },
name: “sdk-repro-stream-unavailable”,
cloud: {
repos: [{ url: “”, startingRef: “development” }],
},
});
const run = await agent.send(“<prompt ~4k chars>”);
console.log(“run”, run.id, run.status);
(async () => {
for await (const e of run.stream()) console.log(“event”, e.type);
})().catch(console.error);
const result = await run.wait();
console.log(“wait”, result.status, result.error, result.durationMs);
console.log(“agentId”, agent.agentId);
Severity
High for automation: first-turn bootstrap cannot complete; agents are not durable after failure.

Add your Cursor team/account email, timestamp (~2026-09-12 UTC), and lockfile @cursor/sdk exact version before sending. I can shorten this to a GitHub-issue template or add a test-agent.ts one-liner repro command if you want.

Operating System

MacOS

Version Information

Version: 3.20.10
VS Code Extension API: 1.128.0
Commit: d6f462cdd0a6a6d1cff570daf980e671d0a63de0
Date: 2026-09-10T21:04:54.045Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 42.10.0
Chromium: 148.0.7778.280
Node.js: 24.18.1
V8: 14.8.178.38-electron.0
xterm.js: 6.1.0-beta.291
OS: Darwin arm64 25.6.0

For AI issues: which model did you use?

composer 2.5

For AI issues: add Request ID with privacy disabled

For Cursor support, cite: agent bc-dbaaba57-e59b-4dcb-bc75-a61ef2aadcea, run run-dd4e3bb8-de6a-4e32-94ed-6a621f422cb0, repo agora-api-monorepo / development, ~4.5k bootstrap prompt.

Additional Information

wasn’t happening few weeks ago, same code

Does this stop you from using Cursor

Yes - Cursor is unusable

mohitjain · September 14, 2026, 6:24am

Hey @Ilan_Levy
None of this is caused by your prompt size or how you pass env/repos. Three separate things:

The runs are failing on our side at environment startup, before your prompt runs. We’ve identified it and are working on it, so there’s no change needed on your end. (A smaller slice is Docker Hub occasionally failing to pull your node:24.12.0-bookworm-slim base image; pinning or mirroring it helps.)
stream_unavailable from run.wait() is a separate known issue, not the server’s verdict on the run. It can fire on runs that actually finish. Treat it as “reconnect and check”: poll GET /v1/agents/{id}/runs/{runId} for the real status instead of trusting wait() alone.
The agent isn’t being deleted. With cloud.envVars, the server assigns the id and agent.agentId only updates after the first agent.send() resolves, so the id captured right after Agent.create() returns 404. Read agent.agentId (or run.agentId) after send() completes and your health checks and Agent.resume will find it. Your real agents are at cursor.com/agents.

We’ve let the team know. If runs are still failing at startup after the next day or so, reply with a fresh agent id captured after send() and I’ll dig back in.

Ilan_Levy · September 20, 2026, 4:48pm

See the follow reproduce script:

/**

 * =============================================================================

 * Cursor SDK repro: create → wait → resume → prompt (repeat)

 * =============================================================================

 *

 * Flow (matches how a long-lived cloud agent is used across turns):

 *

 *   1. Agent.create(named cloud env)     → remember agentId (bc-*)

 *   2. agent.send(prompt) + wait

 *   3. close local handle

 *   4. Agent.resume(agentId)             → same cloud agent by id

 *   5. agent.send(next prompt) + wait

 *   6. repeat resume + send …

 *

 * If resume returns agent_not_found → create a NEW agent and continue.

 *

 * Observed failure (intermittent):

 *   wait.status = error

 *   wait.error.code = stream_unavailable

 *   wait.error.message = "Run stream is no longer available"

 *   durationMs = null, streamedChars = 0

 *

 * Run:

 *   cd myagents-app/test && npm i

 *   CURSOR_API_KEY=… CURSOR_CLOUD=ilan-e2e npm run repro-cloud

 *

 * Exit 2 = a turn hit stream_unavailable (or other run error).

 * Exit 1 = create/resume startup failure we could not recover from.

 * Exit 0 = all planned turns finished.

 * =============================================================================

 */

import { createRequire } from "node:module";

import {

Agent,

CursorAgentError,

type RunResult,

type SDKAgent,

} from "@cursor/sdk";




const require = createRequire(import.meta.url);

const sdkVersion = (

require("./package.json") as { dependencies: { "@cursor/sdk": string } }

).dependencies["@cursor/sdk"];




const STREAM_UNAVAILABLE = "Run stream is no longer available";




const apiKey = process.env.CURSOR_API_KEY?.trim();

const cloudEnvName = process.env.CURSOR_CLOUD?.trim();




if (!apiKey || !cloudEnvName) {

console.error("Need CURSOR_API_KEY and CURSOR_CLOUD=<named-env>");

process.exit(1);

}




const prompts = [

  {

label: "turn-1-bootstrap",

text: [

"Turn 1: bootstrap.",

"Do NOT modify files, commit, or open a PR.",

"Reply with exactly one line: READY turn-1",

    ].join("\n"),

  },

  {

label: "turn-2-investigate",

text: [

"Turn 2: after resume — light investigate.",

"Do NOT modify files, commit, or open a PR.",

"Run: pwd; git rev-parse --abbrev-ref HEAD; ls -la | head -20",

"End with: READY turn-2",

    ].join("\n"),

  },

  {

label: "turn-3-comment",

text: [

"Turn 3: after resume — short comment.",

"Reply with exactly one line: READY turn-3",

    ].join("\n"),

  },

];




type TurnReport = {

label: string;

agentId: string;

runId?: string;

status: string;

errorCode?: string | null;

errorMessage?: string | null;

durationMs?: number | null;

streamedChars: number;

streamUnavailable: boolean;

recreatedAgent: boolean;

};




function isAgentNotFound(err: unknown): boolean {

if (!(err instanceof CursorAgentError)) return false;

const msg = err.message.toLowerCase();

return (

msg.includes("agent_not_found") ||

msg.includes("agent not found") ||

    (err as { code?: string }).code === "agent_not_found"

  );

}




async function createAgent(): Promise<SDKAgent> {

console.log("");

console.log(">>> Agent.create({ cloud.env.name:", cloudEnvName, "})");

const agent = await Agent.create({

apiKey,

model: { id: "composer-2.5" },

name: "sdk-repro-stream-unavailable",

cloud: {

skipReviewerRequest: true,

env: { type: "cloud", name: cloudEnvName },

    },

  });

console.log("<<< created agentId:", agent.agentId);

return agent;

}




/** Docs: resume same bc-* agent by id (cloud auto-detected from prefix). */

async function resumeAgent(agentId: string): Promise<SDKAgent> {

console.log("");

console.log(">>> Agent.resume(", agentId, ")");

const agent = await Agent.resume(agentId, {

apiKey,

model: { id: "composer-2.5" },

cloud: {

skipReviewerRequest: true,

env: { type: "cloud" },

    },

  });

console.log("<<< resumed agentId:", agent.agentId);

return agent;

}




async function closeHandle(agent: SDKAgent): Promise<void> {

// Drop local SDK handle only — do NOT Agent.archive / delete.

// Cloud agent should remain resumable by agentId.

try {

await agent[Symbol.asyncDispose]();

  } catch (err) {

console.warn("close handle warning:", err instanceof Error ? err.message : err);

  }

}




async function sendAndWait(

agent: SDKAgent,

label: string,

prompt: string,

): Promise<{ result: RunResult; streamedChars: number; streamError: unknown }> {

console.log("");

console.log("─".repeat(60));

console.log("SEND", label);

console.log("─".repeat(60));

console.log("agentId:", agent.agentId);

console.log("promptChars:", prompt.length);




const run = await agent.send(prompt);

console.log("runId:", run.id, "initialStatus:", run.status);




const off = run.onDidChangeStatus((s) => console.log("status →", s));




let streamedChars = 0;

let streamError: unknown;

const streamDone = (async () => {

for await (const event of run.stream()) {

console.log("stream event:", event.type);

if (event.type !== "assistant") continue;

for (const block of event.message.content) {

if (block.type === "text") streamedChars += block.text.length;

      }

    }

  })().catch((err) => {

streamError = err;

  });




console.log(">>> wait() …");

const result = await run.wait();

await streamDone;

off();

console.log("<<< wait done");




const err =

result.error && typeof result.error === "object"

      ? (result.error as { message?: string; code?: string })

      : undefined;




console.log("wait.status:", result.status);

console.log("wait.durationMs:", result.durationMs ?? null);

console.log("wait.error.message:", err?.message ?? null);

console.log("wait.error.code:", err?.code ?? null);

console.log("streamedChars:", streamedChars);

console.log(

"streamUnavailable:",

    !!err?.message?.includes(STREAM_UNAVAILABLE),

  );




return { result, streamedChars, streamError };

}




console.log("Cursor SDK repro: create → wait → resume → prompt");

console.log("@cursor/sdk:", sdkVersion);

console.log("node:", process.version);

console.log("cloud.env.name:", cloudEnvName);

console.log("planned turns:", prompts.length);




const reports: TurnReport[] = [];

let agentId: string | undefined;

let failed = false;




try {

// ----- 1) Create -----

let agent = await createAgent();

agentId = agent.agentId;




for (let i = 0; i < prompts.length; i++) {

const step = prompts[i]!;

let recreatedAgent = false;




// ----- 2+) Resume before every turn after the first -----

if (i > 0) {

await closeHandle(agent);

console.log("(local handle closed; cloud agent id kept:", agentId, ")");




try {

agent = await resumeAgent(agentId!);

      } catch (err) {

if (isAgentNotFound(err)) {

console.warn("resume: agent not found — creating a NEW agent");

agent = await createAgent();

agentId = agent.agentId;

recreatedAgent = true;

        } else if (err instanceof CursorAgentError) {

console.error("resume failed:", err.message, "retryable=", err.isRetryable);

process.exit(1);

        } else {

throw err;

        }

      }

    }




// ----- send + wait -----

const { result, streamedChars, streamError } = await sendAndWait(

agent,

step.label,

step.text,

    );




const err =

result.error && typeof result.error === "object"

        ? (result.error as { message?: string; code?: string })

        : undefined;




const streamUnavailable =

      !!err?.message?.includes(STREAM_UNAVAILABLE) ||

      (streamError instanceof Error &&

streamError.message.includes(STREAM_UNAVAILABLE));




reports.push({

label: step.label,

agentId: agent.agentId,

runId: result.id,

status: result.status,

errorCode: err?.code ?? null,

errorMessage: err?.message ?? null,

durationMs: result.durationMs ?? null,

streamedChars,

streamUnavailable,

recreatedAgent,

    });




if (result.status !== "finished") {

failed = true;

console.error("\n*** TURN FAILED — stopping here (useful for support logs) ***");

break;

    }

  }




await closeHandle(agent);

} catch (err) {

if (err instanceof CursorAgentError) {

console.error("startup failed:", err.message, "retryable=", err.isRetryable);

process.exit(1);

  }

throw err;

}




console.log("");

console.log("═".repeat(60));

console.log("SUMMARY (for Cursor support)");

console.log("═".repeat(60));

console.log("finalAgentId:", agentId);

console.log("cloud.env.name:", cloudEnvName);

for (const r of reports) {

console.log(

JSON.stringify({

label: r.label,

agentId: r.agentId,

runId: r.runId,

status: r.status,

errorCode: r.errorCode,

errorMessage: r.errorMessage,

durationMs: r.durationMs,

streamedChars: r.streamedChars,

streamUnavailable: r.streamUnavailable,

recreatedAgent: r.recreatedAgent,

    }),

  );

}

console.log(

"outcome:",

failed ? "FAILURE (see stream_unavailable / error above)" : "ALL TURNS FINISHED",

);




process.exit(failed ? 2 : 0);

Run it with your CURSOR_API_KEY , CURSOR_CLOUD:

CURSOR_API_KEY=crsr_… CURSOR_CLOUD=ilan-e2e npm run repro-cloud

> repro-cloud

> tsx sdk-repro-cloud.ts

Cursor SDK repro: create → wait → resume → prompt

@cursor/sdk: ^1.0.31

node: v22.23.2

cloud.env.name: ilan-e2e

planned turns: 3

>>> Agent.create({ cloud.env.name: ilan-e2e })

<<< created agentId: bc-66ec3097-fd07-4560-8537-f6939377d1bd

────────────────────────────────────────────────────────────

SEND turn-1-bootstrap

────────────────────────────────────────────────────────────

agentId: bc-66ec3097-fd07-4560-8537-f6939377d1bd

promptChars: 103

runId: run-1bce749e-0bfb-43ea-8bb1-0e69a36730f4 initialStatus: running

>>> wait() …

stream event: status

stream event: thinking

stream event: assistant

stream event: status

status → finished

<<< wait done

wait.status: finished

wait.durationMs: 18877

wait.error.message: null

wait.error.code: null

streamedChars: 12

streamUnavailable: false

(local handle closed; cloud agent id kept: bc-66ec3097-fd07-4560-8537-f6939377d1bd )

>>> Agent.resume( bc-66ec3097-fd07-4560-8537-f6939377d1bd )

<<< resumed agentId: bc-66ec3097-fd07-4560-8537-f6939377d1bd

────────────────────────────────────────────────────────────

SEND turn-2-investigate

────────────────────────────────────────────────────────────

agentId: bc-66ec3097-fd07-4560-8537-f6939377d1bd

promptChars: 168

runId: run-da235cc0-1dd0-49fb-8088-fe4a9347bb9f initialStatus: running

>>> wait() …

status → error

<<< wait done

wait.status: error

wait.durationMs: null

wait.error.message: Run stream is no longer available

wait.error.code: stream_unavailable

streamedChars: 0

streamUnavailable: true

*** TURN FAILED — stopping here (useful for support logs) ***

════════════════════════════════════════════════════════════

SUMMARY (for Cursor support)

════════════════════════════════════════════════════════════

finalAgentId: bc-66ec3097-fd07-4560-8537-f6939377d1bd

cloud.env.name: ilan-e2e

{“label”:“turn-1-bootstrap”,“agentId”:“bc-66ec3097-fd07-4560-8537-f6939377d1bd”,“runId”:“run-1bce749e-0bfb-43ea-8bb1-0e69a36730f4”,“status”:“finished”,“errorCode”:null,“errorMessage”:null,“durationMs”:18877,“streamedChars”:12,“streamUnavailable”:false,“recreatedAgent”:false}

{“label”:“turn-2-investigate”,“agentId”:“bc-66ec3097-fd07-4560-8537-f6939377d1bd”,“runId”:“run-da235cc0-1dd0-49fb-8088-fe4a9347bb9f”,“status”:“error”,“errorCode”:“stream_unavailable”,“errorMessage”:“Run stream is no longer available”,“durationMs”:null,“streamedChars”:0,“streamUnavailable”:true,“recreatedAgent”:false}

outcome: FAILURE (see stream_unavailable / error above)

mohitjain · September 21, 2026, 12:16pm

Thanks, that repro isolates it well. That stream_unavailable on the resume turn is the SDK’s stream-attach failing, not the run itself. The run keeps executing on our side and usually finishes fine, so don’t treat it as a failed turn.

So instead of immediately resuming and sending again (that next send() can 409 while the run is still finishing), either re-open the stream for the same runId after a couple of seconds (it replays from the start), or poll GET /v1/agents/{id}/runs/{runId} until it’s FINISHED and read the result there before the next turn.

Still an issue we’re tracking on 1.0.31. No timeline yet, but I’ll post here when there’s an update.

Ilan_Levy · September 21, 2026, 8:47pm

Hi Mohit,
I’ll check allow me few days to reply pls

Ilan_Levy · September 26, 2026, 1:21pm

Still doesn’t work as expected and reproduces even with the workaround

Using the cloud works , but not the Cloud agent thru the cursor sdk

See log after using the polling and wait technique

[orchestrator] [cloud-agent] created AE-50728 agentId=bc-89308760-6de7-4089-81d5-713e33e55ef0
[orchestrator] [AE-50728] bootstrap: skill prompt 5655 chars (~1414 input tok est), extras=0
[orchestrator]
[orchestrator] [bootstrap] sending… agentId=bc-89308760-6de7-4089-81d5-713e33e55ef0
[orchestrator] [bootstrap] shrink-input: 5655 → 5654 chars (−1)
[orchestrator] [bootstrap] run run-45316961-6195-40af-b1e6-9a38d0e7fc23 (running)
[orchestrator] [bootstrap] status → error
[orchestrator] [bootstrap] [support] wait() returned in 775ms status=error durationMs=null streamedChars=0 resultChars=0 errorCode=stream_unavailable errorMessage=Run stream is no longer available streamError=null streamUnavailable=true
[orchestrator] [bootstrap] [stream-recovery] DETECTED stream_unavailable — polling getRun (not re-attaching wait/stream)
[orchestrator] [bootstrap] diagnostics {“runId”:“run-45316961-6195-40af-b1e6-9a38d0e7fc23”,“status”:“error”,“durationMs”:null,“errorMessage”:“Run stream is no longer available”,“errorCode”:“stream_unavailable”,“resultChars”:0,“streamedChars”:0,“streamError”:null,“streamUnavailable”:true,“agentId”:“bc-89308760-6de7-4089-81d5-713e33e55ef0”,“phase”:“pre-recovery”,“errorKeys”:[“message”,“code”]}
[orchestrator] [bootstrap] [stream-recovery] START poll Agent.getRun runId=run-45316961-6195-40af-b1e6-9a38d0e7fc23 agentId=bc-89308760-6de7-4089-81d5-713e33e55ef0 runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #1 elapsed=0ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #1 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #2 elapsed=1750ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #2 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #3 elapsed=3489ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #3 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #4 elapsed=5195ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #4 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #5 elapsed=7089ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #5 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #6 elapsed=8810ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #6 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #7 elapsed=10520ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #7 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #8 elapsed=12352ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #8 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #9 elapsed=14087ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #9 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #10 elapsed=15840ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #10 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #11 elapsed=17893ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #11 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #12 elapsed=19659ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #12 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #13 elapsed=21471ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #13 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #14 elapsed=23209ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #14 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #15 elapsed=24976ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #15 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #16 elapsed=27905ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #16 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #17 elapsed=29617ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #17 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #18 elapsed=31452ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #18 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #19 elapsed=33177ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #19 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #20 elapsed=34888ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #20 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #21 elapsed=36661ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc23) runtime=cloud
[orchestrator] [bootstrap] [stream-recovery] poll #21 getRun failed: [agent_not_found] [agent_not_found] Agent not found
[orchestrator] [bootstrap] [stream-recovery] poll #22 elapsed=38370ms → Agent.getRun(run-45316961-6195-40af-b1e6-9a38d0e7fc

–

Ilan_Levy · October 5, 2026, 3:51pm

Hi @mohitjain Are you able to look at the issue ?

Topic		Replies	Views
Cloud Agents API/SDK stream delivers intermediate messages (steps) ~20–40s later than cursor.com/agents UI on the same run Bug Reports performance , cloud-agents , cursor-sdk	8	158	September 11, 2026
Local @cursor/sdk agent: RUNNING → ERROR [unknown] [internal] (empty conversation); cloud works with same API key Bug Reports cloud-agents , cursor-sdk	1	52	August 29, 2026
Agent.send().wait() returns bare status: "error" while ConnectRPC [unauthenticated] leaks as unhandledRejection Bug Reports cloud-agents , long-running-agents , cursor-sdk	4	173	July 8, 2026
SDK Agent.send streaming runs terminate RUNNING → ERROR with no diagnostic body (@cursor/sdk 1.0.22) Bug Reports cursor-sdk	1	62	August 5, 2026
Cloud-agent SDK runs ERROR in 5-76s — "Workflow execution failed" / shell tool dead Bug Reports terminal , cloud-agents , cursor-sdk	2	110	May 14, 2026

Cloud agent runs fail immediately with stream_unavailable / “Run stream is no longer available”; agent often deleted (404) after first failed bootstrap

Where does the bug appear (feature/product)?

Describe the Bug

Steps to Reproduce

Expected Behavior

Operating System

Version Information

For AI issues: which model did you use?

For AI issues: add Request ID with privacy disabled

Additional Information

Does this stop you from using Cursor

Related topics