Skill Engineering

This page may contain stale information. Last updated: 2026-06-12

Overview

The discipline of designing, optimizing, and deploying reusable agent skill documents for production LLM systems — emerging as a distinct engineering practice alongside prompt engineering and fine-tuning in 2026.

Timeline

• 2026-06-11: microsoft releases skillopt — systematic text-space optimization for .md skill files
• 2026-04-22: google Chrome integrates Gemini skills for browser automation
• 2026-06-11: github Agentic Workflows — markdown-defined CI/CD agent specs

Core Practices

1. Package procedural knowledge in compact markdown (agent-skills)
2. Validate edits against held-out benchmarks (text-space-optimization)
3. Test cross-harness portability (Codex CLI ↔ Claude Code)
4. Optimize for deployment token budget (<2,000 tokens)

Economics

• SkillOpt training: ~$1–5 per single-task skill via GBrain
• No GPU fine-tuning required — API calls to optimizer model only
• One-time training cost amortized at deployment

Related Topics

• enterprise-ai-reliability
• ai-coding-tools
• developer-tools
• coding-agents

Key Tools

• skillopt (MIT, Microsoft Research)
• microsoft-agent-framework
• semantic-kernel

Sources

• 2026-06-11-microsoft-skillopt-open-source
• 2026-06-11-skillopt-github-microsoft