Unverified50% confidenceFactExact time
In practice, stop_reason === 'tool_use' does not reliably reflect message structure: a response may contain a real tool call block yet report end_turn.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedOnly responses with a finish_reason of stop or tool_calls should be cached; responses truncated by length or those that errored should not be cached.69% similarUnverifiedMisreporting tool execution errors as protocol errors causes the model to miss useful hint information and be unable to self-correct, causing the agent to stop or enter an infinite loop.67% similarUnverifiedIt is recommended to write into prompt templates, as a standard tool-calling policy, that execution should stop and be reported after more than N consecutive request failures.66% similarUnverifiedWhen multiple tools return conflicting data, a hard rule should be added to the system prompt strictly prohibiting the model from synthesizing an answer on its own, requiring it to present the conflicting points to the user as-is.64% similarUnverifiedIf the deployment platform's parser cannot correctly recognize tool call tokens and treats them as plain text, the model will enter a waiting state and continuously repeat its reasoning, forming a dead loop.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013814API
curl https://kongchang.com/api/v1/knowledge/claims/1013814MCP
get_claim(id=1013814)