#31VibeQuest — AI Learning Workbench for CKB/Fiber

Problem

AI-assisted coding is becoming a normal way to build, but it creates a dangerous gap for CKB newcomers: a learner can generate code that compiles without understanding the proof boundary it relies on. In CKB, that gap matters because applications often depend on cells, OutPoints, lock scripts, type scripts, witnesses, transaction evidence, and asset constraints. In Fiber flows, the learner also has to reason about invoices, channel state, payment proof, replay risk, and what is checked off-chain versus proven on-chain.

Today, a builder can read docs, ask an AI assistant, or copy snippets, but there is no CKB-native learning loop that forces the builder to inspect generated code, test the denial path, and explain why a proof is safe. VibeQuest fills that gap by treating AI as the execution layer and understanding as the progression layer.

Why CKB

CKB's Cell Model is powerful but unfamiliar to many builders coming from account-based chains. Learning it well requires more than a high-level explanation; builders need to trace concrete verifier code and ask what state is being trusted, where that state came from, and what an attacker could mutate. Lock/type scripts and witnesses make this an ideal fit for proof-of-understanding learning: the learner can inspect a generated verifier, run a denial test, and explain the exact trust boundary.

Fiber adds a second strong CKB-native reason. Payment channels make fast, low-cost learning rewards, paid-access demos, hints, and proof-bound quests practical, while the learner still needs to understand replay prevention and settlement evidence. VibeQuest uses those properties as teaching material rather than abstract theory.

Spec

• Learning profile: the user selects a learning lane, speciality, goal, and pace. Examples include vibecoder, backend developer, frontend developer, auditor, product/community learner, and researcher.
• AI lesson generation: the backend generates a structured multi-module path with deep explanations, code lenses, checkpoints, wrong-answer feedback, tutor context, reference trails, and a bridge into practice.
• Checkpoint gate: the learner must pass a lesson-specific checkpoint before generating the related code quest.
• Lesson-to-quest compiler: the backend turns the completed lesson, checkpoint answer, misconception, concepts, and practice bridge into generated verifier files, denial tests, code explainer, and boss challenge.
• Workbench: the learner reviews generated code, switches files, runs proof/test/denial-path checks, asks code-specific tutor questions, and solves the boss challenge.
• Persistence: MongoDB stores users, learning paths, quest runs, tutor notes, notebook reflections, completion state, and reward claim state.
• Wallet and rewards: JoyID/CCC can bind learner identity, and Fiber invoice-bound reward claims can be prepared after completion checks pass.

Mini-demo

A learner chooses the CKB Cell Model path as a backend developer. VibeQuest generates five modules, beginning with cells as consumed/recreated state. The learner passes a checkpoint about why frontend JSON is not cell evidence. The completed lesson unlocks a quest that generates a verifier and tests. In Workbench, the learner inspects the generated code, runs a denial test that mutates the OutPoint or witness, answers a boss challenge about the trust boundary, and sees the completion recorded in the dashboard.

Demo walkthrough: https://github.com/buidlLabs3/vibequest-web/blob/main/demo.webm

References / prior art

VibeQuest is closest to an AI coding assistant combined with an interactive security lab, but grounded in CKB/Fiber concepts. Instead of only producing code, it makes the learner prove they understand what the generated code trusts, what denial path exists, and where CKB/Fiber evidence enters the application.