@nash_su: Microsoft released something interesting. It uses the idea of training a model to optimize skills, making adjustments to the skill's MD content each time, running the results, and if it improves, keeps pushing forward. This approach is quite interesting.
Summary
Microsoft released a new tool that uses an iterative approach similar to training models to optimize skills. It adjusts the markdown content each time, tests the results, and continuously advances improvements.
View Cached Full Text
Cached at: 06/29/26, 12:27 PM
This thing from Microsoft is kind of interesting.
Using the approach of training models to optimize skills, each time you make some adjustments to the MD content of the skill, then run the results, and if it’s good, continue advancing. This idea is quite interesting.
https://t.co/PVs5eGN1Hd
Similar Articles
@NFTCPS: Microsoft came up with something called SkillOpt, and its approach is pretty wild: treating an agent's skill documentation like a neural network for training, with epochs, batches, learning rates, and validation sets, but without touching a single model weight. What makes it great? Let me break it down into three points: Training only modifies one skill document, and any new changes must be validated on the...
Microsoft introduces SkillOpt, a method that trains an agent's skill documentation like a neural network, using epochs, batches, learning rates, and validation sets for optimization, without modifying model weights. It achieves top results across multiple benchmarks and can be transferred across models and tools.
@FinanceYF5: 2/ SkillOpt: Treating Documents as Trainable Parameters Microsoft treats SKILL.md as trainable model parameters—without changing weights, only optimizing natural language documents, with a validation gate filtering each change. 6 Benchmarks, 52 consecutive wins, GPT-5.5 conversation boost…
Microsoft proposes the SkillOpt method, which treats documents as trainable parameters. By optimizing natural language documents without modifying weights, it improves model performance. It achieves 52 consecutive wins across 6 benchmarks, with GPT-5.5 improving by 23.5 points and Claude Code by 19.1 points.
@Xudong07452910: Let the model set its own problems, solve its own problems, and train itself — the biggest fear is learning incorrect problems along with the correct ones. This paper by the Qwen team proposes Skill Self-Play, adding a continuously updated skill library to the model's self-evolution. There are three roles in training: The Proposer generates tasks that are just challenging enough based on the skills...
The Qwen team proposes the Skill Self-Play framework, which significantly improves model capabilities on tool-calling and reasoning tasks through the collaboration of Proposer, Solver, and a dynamic skill controller in self-play.
@Datou: Microsoft values its reputation, deliberately avoiding synthetic data. They trained a base model using only human data, then split it into three expert models for different domains. They then distilled these three capabilities back into the base model (weight ratio allocation requires experience), followed by a round of reinforcement learning to enable the distilled model to flexibly apply the right capability based on the problem.
Microsoft releases technical details of MAI-Thinking-1 training: uses purely human data to train a base model, then trains three domain expert models, merges capabilities back into the base model via distillation, and then applies reinforcement learning to enable the model to flexibly utilize different capabilities.
@BohuTANG: I used to try cross-model mutual review, but that was too slow for me. Now I've discovered a new method: /harden, which achieves great results with two rounds of convergence on the same model. Interested folks can try this skill.
BohuTANG introduces /harden, a method for same-model two-round convergence, and highlights the evot agent engine which completes complex tasks with fewer tokens and lower cost than alternatives like Claude Code.