My approach to agentic skills

2026-06-20 · Source: Artificial Intelligence in Plain English - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, short

Summary

This article outlines best practices for effective skill management in AI Agentic Engineering, focusing on ensuring agents behave as expected and learn from errors. It emphasizes the importance of carefully vetting external skills, potentially using tools like SkillSpector, and understanding their scope (user or project level). The author advocates automating repetitive tasks by creating single-purpose skills, which are easier to manage and compose into complex workflows. Key recommendations include writing short, effective skill descriptions for efficient triggering and keeping skill content compact to optimize token usage, exemplified by a custom "compact-skill-creator". Crucially, the article highlights a technique for agent self-improvement: instructing agents to update relevant skills or documentation immediately after making a mistake, a process facilitated by a "self-improve" skill.

Key takeaway

For AI Engineers building agentic systems, carefully manage your skills to ensure predictable agent behavior and continuous improvement. Vet all skills, especially external ones, for efficiency and security. Design single-purpose, compact skills with concise descriptions to optimize context windows. Crucially, implement a self-correction mechanism where your agents update their own skills or documentation after making errors, preventing recurrence and enhancing long-term reliability. This approach will significantly sharpen your agent's performance.

Key insights

Effective agentic skill management involves careful design, context optimization, and a feedback loop for self-correction.

Principles

• Scrutinize external skills for hidden instructions or verbosity.
• Design skills with a single, focused purpose.
• Keep skill descriptions and content concise to save tokens.

Method

Implement a feedback mechanism where agents update their own skills or documentation based on identified mistakes, ensuring future correct behavior.

In practice

• Employ tools like SkillSpector for skill vulnerability detection.
• Develop project-agnostic skills, then wrap them for specific project needs.
• Use a "self-improve" skill to automate learning from agent errors.

Topics

• Agentic Engineering
• AI Agents
• Skill Management
• Agent Self-Improvement
• Context Window Optimization
• SkillSpector

Code references

• NVIDIA/SkillSpector
• FrancescoBorzi/agent-toolkit

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Two Ways to Build a Skill Server for Your AI Agent — Dynamically loading skills with rich descriptions improves AI agent tool selection and context window efficiency.
• addyosmani / agent-skills — AI coding agents can achieve production-grade software quality by following structured, verifiable engineering workflows and best practices.
• Skill Issue: How We Used AI to Make Agents Actually Good at Supabase — Pedro Rodrigues, Supabase — Agent skills enhance LLM performance by providing context and workflows through progressive disclosure.
• openai / skills — Agent Skills standardize AI agent capabilities for repeatable task execution across various applications.
• Level-up your AI Agent with Skills Engineering — Skills engineering explicitly guides AI agent decision-making through detailed instructions, improving reliability and performance.
• Anatomy of Agent SKILLS — Agent "skills" prevent context bloat by progressively disclosing information, improving efficiency and modularity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence in Plain English - Medium.