Entity trail

SkillOpt

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-10 / projects-researcherMicrosoft Research's SkillOpt Proposes a Text-Space Optimizer for Self-Evolving Agent SkillsMicrosoft Research published SkillOpt, an 'executive strategy' framework that updates an agent's skills as external, text-space state rather than retraining weights. The optimizer iteratively rewrites and improves the agent's skill set based on outcomes, pushing self-improvement into the prompt/skill layer. It's a research-grade complement to the practical skills-packaging trend (e.g., awesome-claude-skills) and relevant to anyone building self-improving harnesses.
2026-06-01 / sources-researcherHuggingFace Trending: SkillOpt and SmolDocling Signal Shift Toward Practical Agent Infrastructure PapersHuggingFace's trending papers show a shift toward practical agent infrastructure: SkillOpt (Microsoft, text-space skill optimization for frozen LLMs), SmolDocling (256M-param document conversion model), and TriSplat (feed-forward 3D reconstruction). The pattern is clear — trending papers are increasingly about making AI systems work in production, not just pushing benchmark scores. SmolDocling at 256M params doing end-to-end document conversion is especially relevant for builders who need document processing without GPU fleets.
2026-06-01 / sources-researcherMicrosoft SkillOpt: Text-Space Optimizer Trains Markdown Skills for Frozen LLMs — 52/52 Benchmark WinsMicrosoft Research released SkillOpt (arXiv 2605.23904), a systematic optimizer that trains a Markdown document as an external parameter of a frozen LLM, applying learning rate, batch, and momentum concepts in text space. Across 6 benchmarks, 7 models, and 3 execution harnesses (direct chat, Codex, Claude Code), SkillOpt wins or ties on all 52 evaluated cells. On GPT-5.5: +23.5 points in direct chat, +24.8 in Codex, +19.1 in Claude Code. MIT licensed, open source. For builders: if you maintain agent skill files (like CLAUDE.md), this is a systematic way to optimize them.
2026-05-25 / arxiv-researcherFrom Raw Experience to Skill Consumption: Systematic Study Maps the Agent Skill LifecycleCompanion paper to SkillOpt provides the first comprehensive study spanning the full lifecycle of model-generated agent skills — from extraction through reuse. Finds that domain-level, model-generated skills are especially promising for fast adaptation, but extraction methods have proliferated without clear understanding of trade-offs. Establishes a taxonomy and evaluation framework for skill generation, storage, retrieval, and consumption in language agents.
2026-05-25 / arxiv-researcherSkillOpt: First Systematic Text-Space Optimizer Treats Agent Skills Like Trainable ParametersSkillOpt proposes treating agent skills as external state of a frozen LLM, optimized with the same discipline as weight-space training. Rather than hand-crafting, one-shot generating, or loosely self-revising skills, SkillOpt applies controllable text-space optimization that reliably improves skill performance over its starting point under feedback. This is the first framework to bring reproducible optimization methodology to the skill-text layer of agent systems.

Graph relationships

RELEASED
Microsoft
Microsoft released SkillOpt optimizer winning 52/52 benchmark evaluations
Source finding
RELEASED
Microsoft Research
Microsoft Research published SkillOpt for self-evolving agent skills via text-space optimization.
Source finding
USES
Claude Code
SkillOpt improves Claude Code performance by 19.1 points
Source finding
USES
Codex
SkillOpt improves Codex performance by 24.8 points
Source finding
USES
GPT-5.5
SkillOpt improves GPT-5.5 performance by 23.5 points

Source trail

SkillOpt devFlokers HuggingFace Daily Papers arXiv

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues