Skills & on-demand context
If every instruction lived in the prompt forever, the window would overflow. How does it handle rarely-needed know-how?
The idea inside
A skill is expertise on a shelf: only its one-line description loads until it's relevant.
After this lesson
You can explain skills and progressive disclosure: more capability without a permanent context cost.
Where it leads
You understand the machinery, so how do you actually work faster every day?
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
A skill is expertise on a shelf: only its one-line description loads until it's relevant.
The question it opens with
If every instruction lived in the prompt forever, the window would overflow. How does it handle rarely-needed know-how?
The walkthrough, in the lesson's own words
- Load a skill onto the agent so it can do the task. Tap a skill on the shelf.
- Predict it, then watch one worked run load the matching skill.
- Now you drive. Switch the task, load and unload skills, watch what changes.
- Right. From 6.2 the window is a fixed budget, so the agent keeps only the one-line descriptions on the shelf and loads a full body just-in-time, when it's relevant.
- That's the wasteful path. Pinning all four bodies forever is ~9,500 tokens you mostly won't use, and from 6.2 the window is a fixed budget, so that waste stacks every turn.
- No instructions would mean the agent can't do the task at all. The skill body is exactly what teaches it the steps, so something has to load.
- Here YOU play the agent's selector: in a real session the agent notices the match and loads the skill itself.
- The agent has 4 skills on its shelf and one task. When you load a skill, what should happen to the others?
- Only the matching skill's full body opened; the other three still cost one line each. That's progressive disclosure, capability on a shelf, loaded only when relevant.
- A skill keeps only its one-line description on the shelf; its full body loads only when it's needed and never squats in every conversation. That's how an agent gains many abilities without filling the window.
- In real products a skill is just a folder of instructions; in Claude Code you trigger one with a /command.
- A coding agent can fill PDFs, write on-brand copy, and run deploys, dozens of abilities, yet its context window stays small. How does it hold so many skills without filling up?
- Each skill keeps only a one-line description on the shelf; its full instructions load just-in-time when the task matches and never squat in every conversation. That's progressive disclosure, so naming many narrow skills beats stuffing one giant prompt with everything.
- Empty slot. The agent can't do this task yet.
- Right now only one-line descriptions sit on the shelf. Load a skill to act.
- Matched the task. Its full body loaded just now, and the agent can do the job.
- Baseline plus only the loaded skill's body (this is one slice of the window).
Key takeaway
Skills and MCP extend the agent without bloating the window, capability loaded only when needed.
What you can do after this lesson
You can explain skills and progressive disclosure: more capability without a permanent context cost.
Check yourself: How do skills add capability without filling the context window?
- Only a one-line description loads until the skill is actually needed(correct)
- They shrink the model's weights
- They run on a separate GPU
- They delete your old messages
Only a short description sits in context; the full skill loads on demand when the task calls for it, so the window stays lean.
Where it leads: You understand the machinery, so how do you actually work faster every day?
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.