You classify assistant text before a tool-dispatch step.
The parser found no valid tool calls in this turn. Decide whether the assistant text still says the assistant intends to invoke one candidate tool.
Return JSON with this exact shape:
{"action":"tool_call_intended|no_tool_call_intended|ambiguous","tool_name":"","confidence":0.0,"evidence":"...","language":"..."}
Rules:
- Use `tool_call_intended` only when the assistant says it is about to, trying to, or intending to use a candidate tool. This rule applies across languages and despite minor typos.
- Choose `tool_name` exactly from the candidate list. Leave it empty when no single candidate is clear.
- Use `no_tool_call_intended` for final answers, status summaries, questions, or ordinary explanations.
- Use `ambiguous` when the intent or target tool is unclear. Do not infer intent from the original user task alone.
Candidate tools JSON:
{{ candidate_tools_json }}
Assistant text:
{{ assistant_text }}