Learning While the Agent Builds: Learning Opportunities

Advanced section. The setup guides gave your agent a brain and hands, and GrillMe & Superpowers changed how work starts – plan before build. This page adds the missing half: what happens after the work is done. “Learning Opportunities”, a skill by Dr. Cat Hicks, turns finished work into a short exercise so you keep learning while the agent does the typing.

The problem it addresses

Agentic work has a trap the author calls the AI productivity trap: output is fast and fluent, and fluency feels like understanding. You accept generated code, text or analyses without the active processing that actually builds knowledge – and six weeks later you cannot change the thing you “built”. The skill counters exactly this with a few well-studied mechanisms: retrieval practice, prediction before observation, explaining in your own words, deliberate pauses.

What it is

A single skill that changes the agent’s behaviour after substantial work – new files or modules, schema changes, architectural decisions or refactors, unfamiliar patterns, anything where you asked “why” during the work. At that moment the agent offers, briefly and without pressure:

“Would you like to do a quick learning exercise on [topic]? About 10–15 minutes.”

You decide every time. Decline once and it stays quiet for the rest of the session; after two accepted exercises per session it stops offering as well.

The exercises

Exercise	What you do
Prediction → Observation → Reflection	You predict what a piece of the work will do, then you check reality together, then you say what surprised you.
Generation → Comparison	Before seeing the implementation, you sketch how you would approach it – then compare with what was built.
Trace the path	You walk through a concrete case step by step and predict each transition before it is revealed.
Debug this	The agent presents a plausible bug or edge case; you diagnose and fix it before discussing.
Teach it back	You explain a component as if onboarding a new colleague; the agent corrects what is off.
Retrieval check-in	At the start of a follow-up session: what do you still remember from last time?

The key design decision: the agent asks a question and then stops. It waits for your answer – no suggested answers, no hints disguised as encouragement, no answering itself. Wrong guesses are treated as useful data, not failure. This feels unusual at first; the pause is the product. Answering before seeing the answer is precisely what builds retention.

Install on Hermes

Prerequisite: the Hermes setup guide is done. Open a new Hermes session and type (the agent will ask for confirmation before running anything – approve it):

Run this and show me the result:

hermes skills install https://raw.githubusercontent.com/DrCatHicks/learning-opportunities/main/learning-opportunities/skills/learning-opportunities/SKILL.md --yes

(--yes answers the installer’s own confirmation prompt, which cannot be answered inside the agent’s terminal.) Then end the session and start a new one.

Plan B (manual): run the same command yourself in a terminal, without --yes – there you can answer the confirmation prompt directly.

Install on OpenCode

OpenCode has no skill registry, so the agent creates the file directly (approve the file creation):

Create the file ~/.config/opencode/skills/learning-opportunities/SKILL.md with the exact content of
https://raw.githubusercontent.com/DrCatHicks/learning-opportunities/main/learning-opportunities/skills/learning-opportunities/SKILL.md
Do not change the content. Show me the file afterwards.

Then quit OpenCode and start it again.

Plan B (manual): download the file from the URL, create the folder .config/opencode/skills/learning-opportunities/ in your home folder, save the download as SKILL.md inside it, restart OpenCode.

First use

Trigger it honestly: let the agent do a real piece of work – create a small file, change a structure – and wait. The offer should come right after completion.
Accept once: say yes and experience the pause. You can also ask directly: “Do a learning exercise on what we just built.”
Test the brake: decline one offer and check that the agent does not ask again this session.

What you actually gain – honestly

Guided self-study on your own project. The exercises use your real work as the material – no generic tutorial.
You stay the expert. The skill is built so that you generate answers yourself instead of reading the agent’s – that is what separates “I used AI to do this” from “I can do this”.
The price: 10–15 minutes each time, and one interruption after larger pieces of work. If today is pure output day, say no – one “no” buys silence for the session.

Safety: what are you installing?

Skills are instructions the agent will follow – see rule 3 on the safety page. The whole skill is one readable file; look at it before or after installing.

	Learning Opportunities
Source	github.com/DrCatHicks/learning-opportunities
Author	Dr. Cat Hicks, psychological scientist (research on how developers learn; founder of Catharsis Consulting, formerly VP of Research at Pluralsight)
Licence	CC BY 4.0 – free to use, share and adapt, attribution required (unlike the MIT-licensed tools above, keep the attribution when you redistribute modified versions)
Status	public open-source repository, also distributed via the Claude Code and Codex plugin marketplaces, field-tested with several hundred engineers

If you change your mind:

Hermes: hermes skills uninstall learning-opportunities
OpenCode: delete the folder .config/opencode/skills/learning-opportunities/ – gone on the next restart.

If something does not work

Symptom	Cause and fix
No exercise is ever offered	The trigger is substantial work (new files, refactors, design decisions), not every small edit. Create something real – or ask directly: “Do a learning exercise on X”.
The agent answers its own questions or drops hints	That is exactly what the skill forbids – usually a long-session problem: compression washes out the trigger instructions. Start a fresh session. In Hermes, check `hermes skills list` shows learning-opportunities.
OpenCode ignores it after restart	Is the file at `.config/opencode/skills/learning-opportunities/SKILL.md` in your home folder, named exactly `SKILL.md`?
It interrupts too often for your taste	Decline once – the skill then stops offering for the rest of the session. Or uninstall it and reinstall only when you want to study a new project.

Optional extras

Auto mode: the repository also contains learning-opportunities-auto, which offers an exercise after every git commit. Optional; Windows needs a little extra setup – see the README in the repository.
Learning goals: pairs with learning-goal, a companion skill for structured learning goal-setting.
The science: PRINCIPLES.md explains the learning science behind each exercise type – worth reading on its own.
For teams: the repository’s MEASURE-THIS.md playbook offers validated survey items if you want to run the skill as a visible team experiment.