Read on the left. Practice on the right. AI grades what your prompt actually achieves.
Naveo is the gamified platform for getting good at prompt engineering, MCPs, agents and complex flows. Real practice, instant feedback.
First step is free. No card. No install.
Role and context
A good prompt starts by telling the model who it is and what problem it solves. Without that, it hallucinates.
Which of the two prompts will force clean JSON?
B locks the format. The rubric verifies the model doesn't preamble.
Three moves. No video, no slides, no waiting.
- 01
Read on the left
A short page with the principle, the pattern and an anti-pattern. No fifteen minutes of theory before you touch anything.
- 02
Practice on the right
You write the prompt, pick the option, label the anatomy or design the tool schema. The exercise changes with the skill.
- 03
The rubric grades you
We don't compare your prompt to a fixed answer. We grade what your prompt achieves against a rubric with an LLM judge.
We don't grade your prompt. We grade what your prompt achieves.
Ninety percent of what we teach has no single right answer. That's why the heart of the product is a rubric with an LLM judge, not a string comparator.
You are a data extractor. From the text below, return JSON with name, age, email. If a field is missing, leave it as null. Text: "Carlos, no email on file"
Case 3 fails: the model invents a plausible email.
Four tracks. One progression.
You start with the foundational skill and stack up to orchestrating agents.
- 8 lessons · 90 min
Prompt anatomy
Role, context, examples, structured format. The skill everything else stacks on top of.
8 lessonsEnter - 6 lessons · 70 min
Conversations that work
Memory, follow-ups, recovering bad outputs. Driving an LLM multi-turn toward a concrete goal.
6 lessonsEnter - Main course
Tools and MCPs
You design tools the LLM actually decides to use. Your description is what the model reads. Unique content nobody else teaches.
10 lessonsEnter - 8 lessons · 2 h
Complex flows
Chains, routing, agents, evals. When a single prompt isn't enough.
8 lessonsEnter
Each crew member runs a station. Their authority comes from knowing.
Not decoration. Your progress is measured per station: you master Comms with Vega, Engineering with Forge, Security with Hex. Each track unlocks more coordination.
Atlas signs your capstones
Atlas has no station. He crosses all of them. He shows up at the end of each track and, if you pass, he logs it. If not, he sends you to practice with the right person.
- Comms
Vega
The voice that teaches you to structure prompts. Role, context, examples, format. You start hearing her at step one.
prompt-anatomy · few-shot · system vs user - Data
Echo
Conversation memory, surgical corrections, recovering outputs without losing what worked. She doesn't rewrite, she edits.
conversation-goal · surgical-correction - Engineering
Forge
The workshop. Tool schemas, descriptions the LLM understands, MCPs, error handling. Where ideas become tools.
tool-schema · mcp-multi-tool - Mission Ops
Orbit
Chains, routing, agents with a loop. Orchestrates multi-step missions and decides when to parallelize.
chain-design · agent-loop - Security
Hex
Direct and indirect injection, leaks, adversarial validation. If your system touches users, Hex audits it.
prompt-injection · hardening
A short quest every day. From things you've already seen.
Five to eight scenes, two or three minutes, one concept that reinforces your latest lesson. The streak lives here.
- Five to eight mini exercises, not a 30-second quiz.
- Biased to concepts you've already seen. You never get topics you haven't learned yet.
- Daily streak, lives and gems. The ritual compounds.
Surgical correction on a prompt
The user asked for a change. You edit without breaking the rest of the output.
Everything you do is measured. And visible.
XP, streak, lives, gems and mastery per station. Plus the weekly league against everyone else on board.
- T3Vega · Comms82
- T2Echo · Data64
- T1Forge · Engineering41
- T0Hex · Security18
Your first shift starts with Vega.
First step is free and takes five minutes. If you like how it feels, you keep going. If not, you lost nothing.
Logged. Atlas