SkillOpt: Agent skills as trainable parameters

SkillOpt: Agent skills as trainable parameters - Microsoft Research

Skip to main content

Research

Publications Code & data People Microsoft Research blog

Artificial intelligence Audio & acoustics Computer vision Graphics & multimedia Human-computer interaction Human language technologies Search & information retrieval

Data platforms and analytics Hardware & devices Programming languages & software engineering Quantum computing Security, privacy & cryptography Systems & networking

Algorithms Mathematics

Ecology & environment Economics Medical, health & genomics Social sciences Technology for emerging markets

Academic programs Events & academic conferences Microsoft Research Forum

Behind the Tech podcast Microsoft Research blog Microsoft Research Forum Microsoft Research podcast

About Microsoft Research Careers & internships People Emeritus program News & awards Microsoft Research newsletter

Africa AI for Science AI Frontiers Asia-Pacific Cambridge Health Futures India Montreal New England New York City Redmond

Applied Sciences Mixed Reality & AI - Cambridge Mixed Reality & AI - Zurich

Register: Research Forum

Microsoft Security Azure Dynamics 365 Microsoft 365 Microsoft Teams Windows 365

Microsoft AI Azure Space Mixed reality Microsoft HoloLens Microsoft Viva Quantum computing Sustainability

Education Automotive Financial services Government Healthcare Manufacturing Retail

Find a partner Become a partner Partner Network Microsoft Marketplace Software companies

Blog Microsoft Advertising Developer Center Documentation Events Licensing Microsoft Learn Microsoft Research

View Sitemap

Return to Blog Home Microsoft Research Blog

At a glance

AI agents often fail because their instructions, or skills, are manually modified with no guarantee of improvement. SkillOpt turns skill editing into a training process, making agent behavior more reliable without changing model weights.

SkillOpt treats an agent skill file as a trainable parameter outside a frozen target model, turning skill writing from one-shot prompting into a controlled optimization process.

Across six benchmarks, seven target models, and three execution modes, SkillOpt is the best or tied-best method in all 52 evaluation cells, improving performance without updating model weights.

SkillOpt keeps skills compact and auditable through bounded text edits, validation gating, rejected-edit feedback, and slow/meta updates, avoiding uncontrolled prompt drift.

The optimized skills transfer across model scales, agent harnesses, and related tasks, suggesting that they capture reusable workflow knowledge rather than benchmark-specific instructions.

Large language models (LLMs) are increasingly deployed as agents that gather evidence, call tools, and execute multi-step tasks. For these agents, the hard problem is no longer whether they can call a tool, but whether they can complete tasks reliably and consistently. Today, agent skills typically come from three sources: experts write them by hand, a frontier model generates them one-shot, or the agent loosely revises them after execution. None of these approaches behaves like a deep-learning optimizer. They lack step-size control, held-out validation, and any memory of revisions that failed. As a result, skills tend to grow longer and drift with each rewrite, and a revision that seems perfectly reasonable can quietly degrade real task performance. This uncontrolled skill evolution has become a major obstacle on the path from agent prototype to dependable, production-grade deployment.

In our recent paper, SkillOpt: Executive Strategy for Self-Evolving Agent Skills , we reframe the question from “how do we write a better prompt?” to “how do we train the skill?” SkillOpt treats the skill file as a trainable parameter living outside a frozen target model, bringing a training-style optimization loop, consistent gains across 52 evaluation cells, and a compact skill file that stays readable, auditable, and transferable.

Figure 1. A frozen target model executes tasks while a separate optimizer model trains the skill layer from trajectory feedback, exporting the reusable skill file best_ skill.md through validation gating.

How SkillOpt works

Video 1. SkillOpt’s optimization loop, from trajectory collection to the exported skill file.

SkillOpt organizes skill editing as a forward–backward–update cycle in text space. In the forward pass, the frozen target model executes a batch of training tasks with the current skill; the rollout batch size controls how much evidence each update receives. In the backward pass, a separate optimizer model reads the resulting trajectories in reflection minibatches, distilling patterns to preserve from successful trajectories and patterns to correct from failures.

In the update step, the optimizer proposes small add, delete, and replace edits; candidate edits are merged, deduplicated, ranked, and clipped by a textual learning rate—a per-step edit budget. Every candidate skill must then pass a strict validation gate: it is adopted only if it scores strictly higher than the current skill on the held-out validation split. Rejected edits are not discarded; they enter a rejected-edit buffer that serves as negative feedback for later optimizer calls in the same epoch. On a slower cadence, an epoch-wise slow/meta update consolidates longer-horizon lessons that single batches cannot reveal (Figure 2). Together, bounded edits, validation gating, and best-version selection keep skill optimization controllable and auditable, so the skill converges instead of drifting.

Figure 2. The SkillOpt pipeline: trajectory collection, minibatch reflection, bounded text updates, validation gating, and epoch-wise slow/meta updates jointly constrain skill training.

Consistent gains across benchmarks, models, and execution modes

We evaluated SkillOpt across six benchmarks (SearchQA, SpreadsheetBench, OfficeQA, DocVQA, LiveMathematicianBench, and ALFWorld), seven target models from frontier-scale GPT-5.5 to the small open-weight Qwen3.5-4B, and three execution modes (direct chat, Codex, and Claude Code). Counting each combination as one evaluation cell, When measured against human-written skills, one-shot LLM skills, Trace2Skill, TextGrad, GEPA, and EvoSkill, SkillOpt delivered the best or tied for -best results on all 52 cells. These performance improvements are unusually large for a method that updates no model weights. With GPT-5.5 in direct chat, SkillOpt...