Unverified50% confidenceCautionExact time
Contract drift in tool calls (such as a tool changing from returning plain text to returning JSON, or a schema adding a required field) usually does not directly produce an error pointing to the root cause, but instead manifests as the model repeatedly failing to parse, outputting garbled text, or retrying.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTool error messages should be returned to the model as ordinary tool results, containing a plain-language explanation of the problem and what to do, and 'Do not retry' can stop the model from repeatedly retrying hopeless calls.76% similarUnverified工具调用的可靠性高度依赖模型输出严格遵循预定义的 JSON Schema，任何字段类型偏差、缺失或多余都可能导致框架层的解析失败72% similarUnverifiedThe most common model failure is returning something that is 'almost JSON,' such as extra commas, markdown code fences, verbose preambles, or writing the number 3 as the word 'three,' with the AI node reporting success while the workflow crashes only several steps later when parsing fails.72% similarUnverifiedWhen a tool returns an error or empty output, it should be rejected immediately rather than passing on incomplete data — this is a fail-fast handling approach70% similarUnverifiedOnce final JSON parsing fails, subsequent semantic checks are skipped entirely, which is the technical root cause of the 'task completed yet zero score' phenomenon.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038386API
curl https://kongchang.com/api/v1/knowledge/claims/1038386MCP
get_claim(id=1038386)