Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-08-21 · source-backed
An August 20 study of skill induction finds task-level skill extraction often degrades performance below baseline while subtask-level skills raise it, and text-format skills transfer better than code-format ones. arXiv The proposed skill utility score combines specificity and abstractness and predicts transfer without running the task, which makes it a cheap offline audit on a growing skill library. This pairs with last week's result that self-authored skills run 8 to 11 points worse than no skill, and both point at the same thing: skill granularity is a design decision, not an artifact of how you happened to write it down.
Each link below shares sources, entities, or timing with this story.
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (artifact, audit, baseline); pushes against this story (against).
Reported by the same outlet (arxiv.org); overlapping topics (audit, point, transfer); pushes against this story (vs).
Reported by the same outlet (arxiv.org); overlapping topics (agent, baseline, decision); pushes against this story (versus).
Reported by the same outlet (arxiv.org); overlapping topics (agent, baseline, better); pushes against this story (versus).
Same source domain / Shared topic
Reported by the same outlet (arxiv.org); overlapping topics (agent, baseline, skill, worse).
Reported by the same outlet (arxiv.org); overlapping topics (agent, artifact, point, skill).
Shared entity: Subtask / Same source domain / Earlier coverage
Both cover Subtask; reported by the same outlet (arxiv.org); earlier Subtask coverage from 2026-08-10.
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (agent, audit); pushes against this story (against).