Endless sending of "typing" message telegram.

[urtaevS]

Environment & Tools

• PicoClaw Version: (v0.2 1 docker-compose)
• AI Model & Provider: (Kimi via Nvidia)
• Operating System: (Synology)
• Channels: (Telegram)

📸 Steps to Reproduce

1. Start conversation
2. Wait the answer

Seeing endless sending of "typing" message

[nikolasdehor]

Looking at pkg/channels/telegram/telegram.go line 196-222, the StartTyping method starts a goroutine that sends ChatAction(typing) every 4 seconds via a ticker. The stop function (returned cancel) is supposed to terminate this goroutine when the LLM response is ready.

The likely root cause is that the cancel function returned by StartTyping is never called (or called too late) when the provider response fails or times out. Tracing the call path:

In pkg/agent/loop.go, when an LLM call fails or times out, the typing indicator stop function may not be called in every error branch. If the Kimi via Nvidia provider has high latency or returns errors, the typing goroutine keeps running indefinitely because the context (typingCtx) is never canceled.

Specifically, if RunToolLoop panics or the error handling path skips the deferred stopTyping() call, the goroutine will loop forever sending typing actions.

A fix would be to tie the typing goroutine lifecycle to the parent context instead of creating a separate typingCtx. That way, when the agent loop context is canceled (e.g., on timeout), the typing goroutine stops automatically. Alternatively, adding a select on ctx.Done() alongside typingCtx.Done() in the goroutine (line 214) would catch cases where the parent request context expires before the explicit cancel.