openclaw - 💡(How to fix) Fix Native Windows regression on 2026.4.15: Scheduled Task Gateway unhealthy, agent turns hang / return incomplete-turn payloads=0, Telegram replies Agent couldn't generate a response [1 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

I’m seeing a native Windows issue after updating from 2026.4.14 to 2026.4.15.

On 2026.4.15, the Gateway Scheduled Task does not stay healthy, and even when I run the Gateway manually in the foreground, agent turns either hang or end as an incomplete turn with no payloads. Telegram transport itself works, but replies come back as: ⚠️ Agent couldn't generate a response. Please try again.

This appears to be beyond simple config/auth issues, because provider auth probes succeed and Telegram can reach the Gateway.

Environment OS: Windows 10 (Build 26200) OpenClaw version: 2026.4.15 (041266a) Install type: native Windows global install Provider: OpenRouter Default model tested: openrouter/deepseek/deepseek-v3.2 openrouter/auto Location: Petaling Jaya, Malaysia

Steps to reproduce

agents couldnt get a response

Expected behavior

Expected behavior

Gateway Scheduled Task should stay healthy on native Windows Telegram bot should respond normally Local agent smoke tests should either return output or fail with a concrete error, not hang or return incomplete-turn fallback text

Actual behavior

Actual behavior

openclaw gateway status / status --json reports Scheduled Task registered but not running healthy Telegram transport works, but replies are: ⚠️ Agent couldn't generate a response. Please try again. Foreground Gateway can stay alive, but agent turns still fail Local agent smoke test hangs on a blank PowerShell screen Embedded runner logs show incomplete turns with no payloads

OpenClaw version

2026.4.15

Operating system

Windows 11

Install method

npm global

Model

openrouter/deepseek/deepseek-v3.2 openrouter/auto

Provider / routing chain

openclaw > openrouter

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

Things I tested

Config / auth

openclaw config validate passed OpenRouter auth probe succeeded:

Auth probes +----------------------------+------------------------------+------------+ | Model | Profile | Status | +----------------------------+------------------------------+------------+ | openrouter/openrouter/auto | openrouter:default (api_key) | ok · 15.2s |

So this does not look like missing OpenRouter auth.

Gateway / service

The Windows Scheduled Task is registered and points to a real OpenClaw entrypoint: { "service": { "label": "Scheduled Task", "loaded": true, "command": { "programArguments": [ "C:\Program Files\nodejs\node.exe", "C:\Users\limpr\AppData\Roaming\npm\node_modules\openclaw\dist\index.js", "gateway", "--port", "18789" ], "environment": { "OPENCLAW_GATEWAY_PORT": "18789" }, "sourcePath": "C:\Users\limpr\.openclaw\gateway.cmd" }, "runtime": { "status": "stopped", "state": "Ready", "lastRunResult": "267014" } }, "config": { "cli": { "path": "C:\Users\limpr\.openclaw\openclaw.json", "exists": true, "valid": true }, "daemon": { "path": "C:\Users\limpr\.openclaw\openclaw.json", "exists": true, "valid": true } }, "port": { "port": 18789, "status": "free" }, "rpc": { "ok": false, "error": "gateway closed (1006): ", "url": "ws://127.0.0.1:18789" } }

Foreground Gateway

When I run:

openclaw gateway run --verbose

the Gateway can stay alive in the foreground. Startup logs look normal, including Telegram provider startup. That suggests the gateway can run, but agent turns are still broken.

Telegram

Telegram reaches the Gateway, but replies still fail with:

⚠️ Agent couldn't generate a response. Please try again.

Embedded runner log

Foreground logs show:

[agent/embedded] incomplete turn detected: runId=... sessionId=... stopReason=stop payloads=0 — surfacing error to user

Direct model run via infer

I also tested:

openclaw infer model run --model openrouter/auto --prompt "Reply with exactly OPENCLAW-MODEL-OK" --json

and got: { "ok": true, "capability": "model.run", "transport": "local", "provider": "openrouter", "model": "openrouter/auto", "attempts": [], "outputs": [ { "text": "⚠️ Agent couldn't generate a response. Please try again.", "mediaUrl": null } ] } with the same incomplete-turn log:

[agent/embedded] incomplete turn detected ... stopReason=stop payloads=0

Local agent smoke test

I also ran:

openclaw agent --local --agent main --thinking low -m "Reply with exactly WINDOWS-HATCH-OK."

and PowerShell just sat on a blank screen. It did not print output or fail with a clear error.

Additional notes I initially saw config validation issues around old keys like gateway.auth.allowInsecure / credentials, but after cleanup the config validates, so those now look like a red herring rather than the root cause I also tried changing from openrouter/deepseek/deepseek-v3.2 to openrouter/auto, but the same incomplete-turn behavior remained I attempted rolling back to 2026.4.14, but the issue still appeared to persist after that point

My current suspicion This looks like a native Windows runtime/session regression in the embedded agent path, because:

config validates provider auth probe is ok Telegram transport works foreground gateway can run but agent turns either hang or terminate as incomplete turns with payloads=0