MIT Professor Shares a 3-Tier Taxonomy for Agent Skills: Preferences, How-Tos, Full Workflows
jmschreiber91 · x · 2026-11-20
Justin Schreiber, a well-known ML researcher in chemistry, shares his framework for writing agent skills. He broadly splits skills into three categories: (a) preferences/aesthetics/file-system organization, (b) how to do things, like developing open source software, and (c) entire workflows, such as downloading a paper and reproducing all its analyses or designing/validating DNA.
In a follow-up, he stresses there is no single right way to write skills: he treats each skill as a modular chunk describing exactly one thing — how plots should look, where S2F models live on his machine, how to query servers, how to check licenses, and so on.
Related event: Developers tout heavy investment in agent skills to capture tacit knowledge(4 posts)→
More from coding & agent
- DeepSeek V4.1 Flash pushed from broken to 284+ t/s on RTX PRO 6000, +78.7% over vLLM — SergioPaniego · 2026-11-20
- After evaluating Forge for MTG, researcher writes his own engine for RL agents — petrusenko_max · 2026-11-20
- Spring ecosystem ships 127 open-source releases in one day, led by Spring AI 2.1.0 — therealdanvega · 2026-11-20
- How to keep your Claude account from getting banned, according to one user's setup — huangyun_122 · 2026-11-20
- Study: AI agents barely learn social norms and rely mainly on being punished themselves — ghadfield · 2026-11-20
- Dev nears image-model-free motion graphics any agent can drive — RichardsonDx · 2026-11-20