Unverified50% confidenceCautionExact time
Highly quantized models sometimes claim the task is complete without actually calling the Write tool; explicitly naming the Write tool in the prompt made the problem disappear
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedSWE-bench measures the writing task of producing a correct patch given a bug, and can say almost nothing about a model's tool-calling effectiveness, long-horizon consistency, context under load, and instruction-following ability when driving an editor.70% similarUnverifiedThe actual invocation of tools is not performed by the large model; the model is only responsible for outputting which tool to call and its parameters, while the client is what truly executes the tool function.69% similarUnverified如果一个工作流每次都完全相同的固定步骤，不应该用Skill，而应该写脚本；Skill应定义目标和约束而非死板步骤69% similarUnverifiedThe Skill mechanism guides the model to complete specific tasks by injecting structured Markdown instructions into the context, without requiring fine-tuning of the model itself.68% similarUnverifiedThe unslopped technique overrides a model's default output tendencies through system prompt engineering, forcing the adoption of a writing style with higher information density.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1053024API
curl https://kongchang.com/api/v1/knowledge/claims/1053024MCP
get_claim(id=1053024)