Papers Reveal When Agent Skill Retrieval Helps—and Hurts

New papers show skill retrieval can hurt underlying task performance in some LLMs, while distilling experience into SKILL.md files beats workflow memory by 6 points.

2026-09-08 ~ 2026-09-08 · 4 related posts

• Distilled SKILL.md beats Workflow Memory by 6.06 points — skills stabilize execution, not knowledge — rohanpaul_ai · 2026-09-08
• arXiv: 8,135-trial study shows agent skills stabilize execution, retrieval precision drops to 3.3% — rohanpaul_ai · 2026-09-08
• Paper: skill retrieval gains can hide worse performance on the very tasks retrieved — rohanpaul_ai · 2026-09-08
• Skill Following paper: LLMs with skill retrieval often hurt the very tasks they retrieve for — rohanpaul_ai · 2026-09-08