The AI Daily Brief: How to Use Agent Skills

Host: Nathaniel Whittemore (NLW)
Date: March 18, 2026

Episode Overview

This episode of The AI Daily Brief takes a practical, hands-on look at the growing importance of "skills" in the evolving agentic era of AI. Inspired by a post from Tariq at Anthropic’s Claude Code team, NLW explores how AI agents are being equipped with modular skills to deliver robust, reusable solutions—from the cutting edge of coding automation to mainstream productivity tools. The episode breaks down what agent skills are, why they matter, key best practices for building and using skills, and how this paradigm is rapidly becoming foundational across the AI landscape.

Key Discussion Points and Insights

1. What Are Agent Skills? [29:45]

• Definition:
  Agent skills are "folders of instructions, scripts, and resources that agents can discover and use to perform better at specific tasks. Write once, use everywhere."
• Background:
  As coding agents advanced, developers faced ballooning system prompts and crowded context windows, leading to slow, expensive, and unreliable agents.
• Solution:
  Instead of packing all knowledge at once, skills allow agents to “load the right knowledge at the right moment.”

2. The Anatomy & Progressive Disclosure of Skills [32:00]

• Skill Structure:
  • Each skill is a directory, anchored by a Skill.md file with metadata (name, description) and relevant scripts or resources.
  • Progressive disclosure:
    • Quick descriptions for agent awareness.
    • Detailed Skill.md content when needed.
    • Linked scripts or files for deeper, context-specific use.
• Not Just Markdown:
  "A common misconception…is that they are just markdown files. But…the most interesting part of skills is that they're not just text files. They're folders that can include scripts, assets, data, etc. that the agent can discover, explore and manipulate." – (paraphrasing Tariq, [34:20])

3. Widespread Adoption and Use Cases [35:12]

• Ecosystem Momentum:
  Skills have expanded beyond Anthropic, now supported by OpenAI, Copilot, and many other platforms.
• Common Needs:
  28,000+ skills cataloged on “Clawhub”; common examples include tools, file format handling, and workflow automation.

4. Anthropic’s Taxonomy of Skills [36:50]

• Nine Categories (selected highlights):
  • Data Fetching & Analysis: Connect to data sources, fetch dashboards, common data workflows.
  • Business Process & Team Automation: Automate repetitive workflows (e.g. weekly recap posts).
  • Code Quality & Review: Agents enforce code quality, run adversarial reviews, style checks—crucial as code volume explodes.
  • Verification: High-ROI category—skills to test or verify code, e.g., “having Claude record a video of its output so you can see exactly what is tested.” ([39:00])

5. Best Practices for Building Skills [41:20]

Drawing from Tariq’s Anthropic post:

• Skill Creator Tool Enhancements:
  Helps authors write, test, benchmark, and maintain skills as models evolve—no coding required.
• Key Upgrades:
  • Evals/benchmarks for skills.
  • AB tests for model changes.
  • Auto-rewriting skill descriptions for better triggering.
• Ollie Lemon notes:
  “Anthropic shipped three upgrades to skills that fix most problems… Now you can run evals, skills break less when models update, and descriptions trigger more reliably.” ([42:20])

6. “Capability Uplift” vs. “Encoded Preference” Skills [43:00]

• Capability Uplift: Skills that add fundamentally new functions to the agent.
• Encoded Preference: Skills that sequence tasks according to team workflows (more durable, workflow-aligned).

7. Top Tips for Skill Creators [44:00]

• Don’t State the Obvious: Focus on information that extends or corrects base AI knowledge.
• Build a ‘Gotcha’ Section:
  “The highest signal content in any skill is the gotcha section…articulate common failure points…and update over time.” – (paraphrased, Tariq [45:10])
• Think in Terms of the File System: Use full folder structure/context, not just markdown.
• Avoid Railroading the Agent: Provide flexibility, not rigid step-by-step rails.

8. Skills—For Whom? [47:00]

• Advanced Agent Builders:
  Use skills as modular, sharable expertise for large agent teams.
• Power Users:
  Skills = reusable prompts with superpowers. Package instructions, code templates, and references for tasks you do regularly.
• Mainstream Users:
  Even in consumer tools (Notion AI, Perplexity), “skills” are showing up as reusable commands—“write a prompt, you’ll use it once; write a skill and you’ll use it forever.” (Notion AI, [48:30])

9. Emerging Patterns and Broader Implications

• Skills mark the shift from one-off chats to libraries of repeatable, reliable AI capabilities.
• “Even if you’re not an agent builder…the shift is from thinking about ad hoc prompting to reusable capabilities.”
• “The underlying idea is that AI is less and less a one-off conversation and more and more a library of reliable, repeatable capabilities.”

Notable Quotes & Memorable Moments

• On Agent Mobility:
  "It feels pretty magical to give Claude a mission on my computer and get occasional updates…" – Felik Risberg ([03:40])
• On Security:
  “Claude cowork dispatch covers 90% of what I was trying to use OpenClaw for, but feels far less likely to upload my entire drive to a malware site.” – Ethan Mollick ([05:16])
• On Skills Structure:
  “A common misconception…they are just markdown files. But…the most interesting part…is that they're not just text files. They're folders…” – paraphrasing Tariq ([34:20])
• On the Future of Code Review:
  "Even if it would be better if all code…had human review, I don't think there's any chance that that paradigm gets out of 2026." – NLW ([38:22])
• On ‘Gotcha Sections’:
  “The highest signal content in any skill is the gotcha section. These sections articulate common failure points that Claude runs into when using your skill, and ideally…you update your skill over time to capture these gotchas.” – Tariq ([45:10])
• On Rethinking Prompting:
  “Write a prompt, you’ll use it once; write a skill and you’ll use it forever.” – Notion AI ([48:30])

Timestamps for Important Segments

• [00:45] — Claude Cowork’s Dispatch Feature & Industry Implications
• [13:00] — OpenClaw’s Rise in China and Regulatory Reaction
• [18:30] — Meta/Manus Regulatory Scrutiny in China
• [23:40] — Nvidia China Exports & Amazon’s AI Forecasts
• [29:45] — Main Topic: What Are Agent Skills?
• [32:00] — Anatomy & Progressive Disclosure of Skills
• [36:50] — Anthropic’s Skills Taxonomy
• [41:20] — Anthropic’s Best Practices for Skill Builders
• [43:00] — Capability Uplift vs. Encoded Preference Skills
• [44:00] — Anthropic’s Top Tips for Making Skills
• [47:00] — Use Case Tiers: Builders, Power Users, Mainstream
• [48:30] — Skills in Consumer-Oriented Platforms (Notion AI Example)

Conclusion

NLW’s episode underscores that “skills” are not just a new buzzword, but a transformative building block for the agentic AI era—bridging the gap between prompt engineering and true modular expertise. Whether you’re a developer, power user, or an everyday knowledge worker, understanding and leveraging agent skills will soon become second nature as agents move from novelty helpers to integral work companions. As NLW puts it, "AI is less and less a one-off conversation and more and more a library of reliable, repeatable capabilities.”