Stops without returning any log or error

[alexsiu398]

Machine:linux-22.04.4
model:gemini-2.5-flash
Issue:I prompted a task and it is on thinking process but it stops without returning any log or error. I don't know whether it is related to llm api or framework error.

console:
I will use python to parse the content
and extract the data.

continue

Executing Python code...

I apologize for the previous
error. I forgot to include the
html_content within the python script
itself. I will correct that and also
refine the parsing logic to extract the
data more accurately from the provided
content list. I will iterate through
the content list and identify the
tr elements that represent each
cryptocurrency row. Then, I will
extract the relevant data points from
each row.

[norika1207-lab]

This kind of silent stop should probably become an explicit terminal state in the framework, even if the root cause is the upstream model/API.

From an operator point of view, “no log and no error” is worse than a failed tool call because there is no recovery handle. I would capture at least:

• last model response status / finish reason
• whether a tool call was emitted, parsed, rejected, or started
• tool execution start/end/error timestamps
• last streamed token time
• framework state when the turn ended: waiting_for_model, waiting_for_tool, waiting_for_user, completed, failed

Then continue can decide whether it is resuming a failed tool step, asking the model again, or starting a new user turn. Without that state, the agent can accidentally treat an incomplete execution as a normal pause.