I Built Interactive CLI Drills — Learners Quit at Prompt #3

I built what I wanted as a learner: interactive CLI drills in the browser. Type commands. Get feedback. No “install fifteen tools first.” Then the analytics (and the hallway conversations) said something rude: a chunk of learners quit at prompt #3.

This is a show-and-tell about that product bruise—and what I changed.

What I shipped first

• Short missions for CMD, PowerShell, and Bash-oriented tasks on Windows-friendly paths.
• Immediate correctness checks.
• A clean terminal UI that feels like practice, not a MOOC video.

The thesis: skill is built at the prompt. Screenshots weren’t enough (I had already burned myself on that experiment).

Prompt #3: the cliff

Patterns in the drop-off:

1. Prompt #1 was a win (list files, print a string). Dopamine.
2. Prompt #2 introduced one new concept. Still okay.
3. Prompt #3 combined concepts—or required recovery from error—without a scaffold. Confidence cratered.

It wasn’t “students are lazy.” It was difficulty curve arrogance. I had optimized for correctness of the skill model, not for the emotional runway of a first session.

Product fixes that mattered

Smaller concept steps

One idea per prompt until the streak feels earned. Combine later in a “mission boss,” not at step three.

Intentional recovery prompts

Instead of only happy paths, add a safe failure that teaches ERRORLEVEL / $LASTEXITCODE early—after a success, not instead of one.

Shell labeling everywhere

Ambiguity across CMD vs PowerShell was silent quit fuel. Labels and prompts now scream which contract you’re in.

Softer first-session definition of done

A complete first session can be six tiny wins. Heroics can be session two.

What I’d still defend

Interactive > passive for CLI. The quit data didn’t kill the thesis; it killed my pacing ego.

Also: soft CTAs beat launch-speak. The product exists to train operators, not to win a landing-page award.

Where it lives

Instrumentation that changed my mind

I stopped celebrating completion rate alone. I started watching:

• median time-to-first-success,
• drop-off by prompt index,
• repeat visits within 48 hours,
• qualitative “I felt dumb” comments (painful, gold).

Prompt #3 was the cliff on the first curriculum graph. After re-pacing, the cliff moved later—and got shorter. That’s product work, not vibes.

What “showdev” means here

Not a launch. A build diary: interactive Windows CLI practice is the medium; curriculum pacing is the hard problem; soft CTA is intentional. If you’re building a trainer, steal the quit-index chart idea before you polish marketing copy.

Design smell: “clever” early combines

If prompt #3 needs two new ideas and perfect quoting and an error recovery, you didn’t write a drill—you wrote a hazing ritual. Split it. Clever can wait until the streak feels real.

What I almost did wrong next

I almost responded to quit data by adding badges and streaks—gamification as a bandage. Maybe later. First fix was honesty about pacing. If the third prompt is a cliff, glitter won’t pave it. Make the third prompt a fair win, then earn the harder combines.

The drills are free and browser-based: CMD Master—interactive terminal learning for CMD, PowerShell, and Bash. If you try it, watch where you want to bail. That spot is probably my next design ticket.

Open build question

For those who’ve built coding or ops trainers: do you gate progress strictly, or let learners skip and return? I’m still torn between “force the reps” and “don’t imprison adults.”

Show-and-tell, not victory lap: prompt #3 humbled the roadmap.