Summary of 1003: Skills Skills Skills

Overview of Syntax episode 1003: Skills Skills Skills

In this episode of Syntax, Wes Bos and Scott Tolinski dig into the growing world of “skills” for AI agents: lightweight Markdown-based instructions that teach an agent how to handle specific workflows, enforce preferences, and avoid common mistakes. They each share several of their favorite skills—many of which they built themselves—and discuss where skills are genuinely useful versus where they become hype, especially around design, copywriting, and “AI-generated slop.”

Skills Wes and Scott Talked About

Wes Bos’s skills

• Hot Tip Skill

  • Turns his social/video tip workflow into a repeatable Markdown-driven process.
  • Helps gather content, collect post URLs from multiple platforms, and generate a structured tip page with front matter and links.
  • Replaces brittle NPM/pipeline scripts with a flexible agent workflow.

• CSS Motion Systems

  • Gives the AI clear rules for motion design in CSS.
  • Covers:
    • using linear() and better easing approaches
    • respecting prefers-reduced-motion
    • animating performant properties
    • avoiding overused/cheap-looking effects
  • Also includes guidance on using view transitions meaningfully and not applying hover/scale effects everywhere.

• Agent Browser

  • Instructs an agent how to use a headless or connected browser.
  • Can work with an existing Chrome instance or Electron apps via the Chrome debugging protocol.
  • Useful for debugging, scraping, screenshots, and interacting with apps like Slack.

• HTML Skill

  • Sets rules for writing semantic, clean HTML.
  • Pushes back on AI habits like:
    • excessive div soup
    • wrapping everything in lists unnecessarily
    • treating buttons like links
    • adding redundant attributes or accessibility/security patterns that are no longer needed
  • Meant to curb common AI HTML anti-patterns.

Scott Tolinski’s skills

• Extract Logos

  • Helps agents find and download the right SVG logo for software/projects/companies.
  • Pulls from sources like SVGL and icon packs.
  • Prevents the AI from substituting random Unicode or wrong logo approximations.

• Dex Task Skill

  • Uses Dex for task tracking between agents.
  • Defines how to:
    • create and update tasks
    • link related tasks
    • manage parent/child relationships
    • determine the next task to work on
  • Scott likes Dex because it’s just a JSON file plus CLI, making it portable and team-friendly.

• Remotion / Hyperframe Skills

  • Guide agents in programmatically generating video.
  • Focus on technical best practices for audio, animation, timing, and composition.
  • Useful for slide/video generation, but not a substitute for good taste or design skill.

• Marketing Skills / Copywriting Skill

  • A skill bundle from Corey Haynes focused on marketing and writing.
  • Scott mainly values the copywriting part for improving:
    • specificity
    • clarity
    • CTA quality
    • landing page messaging
  • He notes it can help make writing more effective, but worries about overuse and AI-generated fake-sounding text.

Main Takeaways

• Skills are best for process, not magic.

  • They work well when they encode a workflow, technical constraint, or house style.
  • They are much less useful when they’re expected to replace taste, judgment, or creativity.

• Technical skills are the strongest use case.

  • Browser automation, task management, HTML rules, motion systems, and media-generation workflows are all good candidates.
  • These are areas where consistency matters and the AI can benefit from explicit guidance.

• Taste still matters.

  • Both hosts warn against skills that simply produce generic, overused, “AI aesthetic” output.
  • A skill can make output more correct, but not automatically more original or beautiful.

• AI-written marketing/copy is controversial.

  • Scott sees it as useful for sharpening and structuring ideas.
  • Wes is skeptical of using AI to generate persuasive text that sounds fake or generic.
  • They agree that poorly written AI text is easy to spot and often weakens trust.

• Skills can be generated from good conversations.

  • If a chat leads to a solid process or set of rules, you can distill that conversation into a skill file.
  • Several skills in the ecosystem are effectively formalized versions of “what worked in the chat.”

Notable Insights

• “Skills” are essentially Markdown documents that define agent behavior.
• Many of the best skills are just codified expertise and preferences.
• The hosts are wary of the “grift” side of the skill trend—paying for something that should probably just be a text file.
• They like skills that help agents stay consistent, semantic, and technically accurate.
• They dislike anything that encourages generic output, bloated HTML, fake marketing language, or over-styled design.

Practical Recommendations

• Use skills for:
  • repeatable technical workflows
  • enforcing code/style conventions
  • task tracking
  • browser automation
  • media generation pipelines
• Avoid using skills to:
  • auto-generate entire blog posts or marketing pages without review
  • flatten every project into the same aesthetic
  • replace genuine writing or design judgment
• Consider turning successful agent chats into reusable skills.
• Prefer skills that are:
  • small and focused
  • easy to maintain
  • committed to a repo
  • useful to your team, not just your local setup

Final Thought

The episode is essentially a practical tour of how Wes and Scott use AI agent skills in their day-to-day development work. Their core message: skills are powerful when they encode expertise and workflow—but they should support good engineering and good taste, not replace them.