openclaw - 💡(How to fix) Fix [Bug]: Tool calls not executed in Telegram direct session (exec/browser always KO) on 2026.3.24 [2 comments, 1 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

In Telegram direct sessions on OpenClaw 2026.3.24, the agent replies in chat but does not execute tools (exec/browser consistently fail).

Steps to reproduce

1. Start OpenClaw 2026.3.24 on macOS (gateway running, Telegram connected).
2. In Telegram DM with the bot, send: Test exec: esegui date.
3. In Telegram DM, send: Test browser: apri https://example.com e fai screenshot.
4. Observe no real tool execution; only text responses indicating action not executed / KO.

Expected behavior

The agent should execute tool calls: exec should return current date output, and browser should open the page and return a screenshot.

Actual behavior

The agent returns textual replies but tool execution does not happen (exec/browser always KO). Logs also show repeated SIGTERM/restart cycles on gateway, including "signal SIGTERM received" and "shutdown timed out; exiting without full cleanup".

OpenClaw version

2026.3.24

Operating system

MacOS 26.4

Install method

Homebrew + npm global install on macOS, launched as LaunchAgent (ai.openclaw.gateway) and managed via openclaw CLI.

Model

openai-codex/gpt-5.3-codex

Provider / routing chain

Telegram client -> OpenClaw Gateway (local LaunchAgent on macOS, ws://127.0.0.1:18789) -> OpenAI Codex provider (openai-codex/gpt-5.3-codex via oauth). No external reverse proxy in front of gateway.

Additional provider/model setup details

No response

Logs, screenshots, and evidence

### Evidence

#### User-visible behavior
- In Telegram DM, agent replies textually but does not execute tools.
- Repro commands:
- `Test exec: esegui date` -> response indicates action not executed / KO.
- `Test browser: apri https://example.com e fai screenshot` -> response indicates action not executed / KO.

#### Gateway / runtime logs (key lines observed)
- `signal SIGTERM received`
- `received SIGTERM; shutting down`
- `shutdown timed out; exiting without full cleanup`
- repeated restart cycles with new PIDs
- `gateway already running under launchd; waiting 5000ms before retrying startup`

#### launchctl evidence
`launchctl print gui/501/ai.openclaw.gateway` showed:
- `state = running`
- `last terminating signal = Terminated: 15`

#### Attempts already made
- `/start` in Telegram
- `openclaw gateway restart`
- `openclaw doctor --fix`
- disabled backup LaunchAgent (`com.stormbot.backup-openclaw`)
- full machine reboot
- issue persists

Impact and severity

Affected users/systems/channels: observed in my production setup on macOS using Telegram direct chat with OpenClaw main agent (session agent:main:telegram:direct:43557369). Severity: workflow-blocking for operational tasks, because tool execution (exec/browser) does not occur. Frequency: frequent and reproducible during the incident window; repeated KO on every direct tool test. Consequence: automation tasks cannot be completed from chat (no command execution, no screenshots), causing repeated manual troubleshooting, time loss, and inability to deliver scheduled/reporting workflows.

Additional information

No response