I Made a Claude Code Plugin That Forces You to Learn

Summary of the subtitles (tech concepts, product features, and analysis)

Core problem with learning to code in the AI era

• Beginners increasingly rely on AI to generate projects quickly, which raises the bar for what counts as a “good” third-party project.
• Traditional curricula don’t update fast enough, so students often lack the foundation to build quality software without constant help from neural networks.
• Auto-generated code makes it harder to tell who truly understands programming.

Plugin introduced: “Vibewise” (Claude Code plugin)

• Purpose: Force learner participation in the software design/specification process.
• Contrast to typical AI behavior:
  • Usual AI workflow: choose from suggested options (often “A/B/C”), then the AI generates code end-to-end.
  • Vibewise workflow: the AI does not write code automatically. Instead, it repeatedly asks the user questions so they must decide architecture, features, and behavior.
• How it prevents “hands-off” coding:
  • Rather than generating database tables directly, it asks what tables are needed and/or which database the user would choose.
• Overall approach: The AI acts as a guided design partner rather than a code writer.

Demonstration: “Regular Claude” vs “Vibewise”

Regular Claude (baseline)

• The creator provides minimal context, e.g., “I need to make a Notion clone.”
• Claude asks a few high-level questions and recommends options (e.g., “block editor or pages,” plus a recommended stack).
• It then proposes assumptions such as:
  • one user
  • no login
  • no backend (frontend-only)
  • includes frontend features like:
    • page tree
    • unlimited nesting
    • multiple block types
    • searching
    • storage
• The creator receives:
  • an 11-step implementation plan
  • plus a large bundle of architecture/details (data model, storage layer, routing, editor, styling)
• Creator critique:
  • This can overwhelm learners and doesn’t teach effectively.
  • Reverse engineering thousands of lines isn’t intuitive, and learners may not internalize the underlying choices.

Vibewise (learning-focused flow)

• After installing the plugin, the creator repeats a similar goal: “study” / “make an analogue of Notion.”
• The plugin asks structured questions, including:
  • overall programming experience level (e.g., “I am a newbie”)
  • preferred learning cycle (e.g., “standard one,” ask questions as you go)
• It provides a progress/profiling/map of the project as it evolves.
• It uses checkpoints to force scoping and design decisions:

1. Build checkpoint (scope definition)

   • Requires specifying what the app should do when opened (example: open app → log in → create note → start editing).
   • Uses free-response questions to avoid multiple-choice autopilot.

2. Assembly/checkpoint for behavior

   • Forces detailed questions and edge cases, such as:
     • what happens after returning later (cross-device persistence)
     • what “editing” means (block types like headings, lists, checkboxes)

3. Folders and data behavior

   • Asks about folder rules and constraints:
     • can folders contain other folders?
     • can notes exist outside folders?
     • deletion behavior
   • Emphasis: these “small” decisions matter but are often abstracted away by AI.

• Overall goal: transform “vibe coding” into a real specification and development process, where the user designs system behavior and constraints.

Why this is positioned as better for learning

• Real software development includes many small engineering decisions and edge cases.
• AI that abstracts everything can produce a working result, but may reduce actual understanding.
• Vibewise keeps the user responsible for specifying/confirming key system behaviors before any implementation.

Checkpoint model (phases)

• The plugin’s README is referenced as having three phases/checkpoints:
  • Creation phase
  • Design phase
  • Implementation phase
• Implementation happens only after design development and finalization are completed holistically.
• The user participates in design and evaluation (including nuances and possible problems) before code is generated/written.

Call to action

• The video description includes a link to the plugin.
• Viewers are encouraged to file issue tickets or leave feedback in comments so updates are easier.

Main speaker / sources

• Main speaker: The video creator/host demonstrating the Claude plugin and narrating the comparison between “regular Claude” and Vibewise.
• Source system/software referenced: Claude (AI) and the Vibewise Claude Code plugin.