Agent Skills and CLI
Focuswhat skills are and how SKILL.md plus scripts encode a workflow; web skills and extending agent capability beyond the repository; working effectively from the CLI.
You buildA packaged skill with a helper script, plus a browser-driven web skill
Core material≈ 3 h 40 min
- 1Course unit≈ 60 minAgent SkillsHugging Face
The best structured introduction to portable, progressively disclosed skills.
- 2Article20 minEquipping agents for the real world with Agent SkillsAnthropic
, with the public skills repository as worked examples.
- 3
- 4Video45 minThe Beginner's Guide to Coding with CursorLee Robinson of Cursor
Typed languages, linting, formatting, tests, branch review, and parallel background work.
- 5CLI foundation≈ 60 min with exercisesThe Missing Semester 2026: Course Overview and Introduction to the ShellMIT
Shell navigation, composition, scripting, streams, permissions, and safe Bash practices.
- 6Standard10 minAGENTS.md
Read the examples and conventions now; repository instructions become the main topic in Week 4.
Tools and references
- LinkCommand Line Interface Guidelines
How to design CLIs that agents (and humans) can drive reliably. Apply it to your skill's helper script.
- LinkCommon workflowsClaude Code
Git worktrees, parallel sessions, and resuming conversations from the CLI.
Additional video track
- Hands-on skills workshop1 h 19 minSkill Issue: How We Used AI to Make Agents Actually Good at SupabasePedro Rodrigues of Supabase
Builds and evaluates
SKILL.md, supporting scripts, progressive disclosure, and routing behavior. - Web skills18 minBringing Agents onto the World Wide WebPaul Klein of Browserbase
Browser-agent harnesses, network interception, WebMCP, Playwright CLI, memory, and reusable website skills.
- Long form2 h 29 minFuture of Programming with AILex Fridman with the Cursor team
Editor design, speculative edits, and how the Cursor founders think about agents.
Build
Find a workflow you have repeated at least three times: release notes, database migration checks, API endpoint scaffolding, dependency upgrades, or UI regression checks. Package it as a skill with:
- A short
SKILL.mdthat tells the agent when to use it. - Progressive disclosure: keep rare details in linked references.
- One deterministic helper script that follows the CLI guidelines above.
- A fixture or sample input.
- A pass/fail verification command.
Run it from a fresh agent session on two different inputs. Record where the instructions were ambiguous or overloaded the context. Then add one web skill: a browser-driven check (Playwright MCP or Chrome DevTools MCP) that verifies something in your running app.