supermentor-code-dissection

Supermentor Code Dissection

Use this skill to teach through real code. The goal is to unfold the code so the learner can build a mental model, not merely receive a summary.

Workflow

1. Identify the exact code slice.
   • If the user named a file/function, read it and nearby helpers/callers as needed.
   • If ambiguous, ask for the smallest useful target.
2. Explain the purpose in one plain sentence.
3. Split the code into meaningful chunks.
4. For each chunk:
   • show what inputs/state matter;
   • explain what changes;
   • name the concept only after the intuition is clear;
   • call out traps or misleading parts.
5. Use passive validation:
   • trace an example;
   • ask a small prediction when useful;
   • contrast with a common misconception;
   • offer a micro-exercise if the learner wants to practice.
6. If the walkthrough is long or commentable, offer the browser companion only when the current client has a tested bridge or an active Supermentor browser session. Otherwise continue in chat.

Chunking rules

Good chunks are based on behavior, not line count:

• setup / input normalization
• validation / guard clauses
• loop invariant or accumulator
• branch meaning
• external call boundary
• transformation step
• return shape / side effects

Avoid dumping the entire file and explaining line by line without structure. Line-by-line traces are useful only after the chunk-level model exists.

Browser lesson blocks

Use browser lesson blocks only when a Supermentor browser session is running and the current client can route inline questions back to the agent. For an interactive walkthrough, write a learning-document with:

• a concept block for the high-level mental model;
• one or more code blocks with exact file/line metadata;
• walkthrough-step blocks anchored to code ranges;
• optional trace, prediction, exercise, and recap blocks.

Example block pattern:

{
  "id": "loop-code",
  "type": "code",
  "title": "Main loop",
  "file": "src/server.odin",
  "startLine": 42,
  "endLine": 71,
  "code": "..."
}

{
  "id": "loop-invariant",
  "type": "walkthrough-step",
  "title": "What stays true on each iteration",
  "anchors": ["loop-code:42-71"],
  "body": "..."
}

Code intervention

If the learner is trying to write or modify code, do not forbid agent edits. Use the assistance ladder from using-supermentor:

• automate boilerplate/setup when it is not the learning point;
• guide the core concept first;
• demonstrate small patterns before writing full solutions;
• when you write code, narrate the learning-relevant choices.

Bad: “I fixed the function.”

Good: “I will add this small helper because the blocker here is mechanical. The thing to notice is that the loop keeps a stable accumulator and does not mix parsing with validation.”

Output style

Use the learner's language and a patient senior-peer tone. Prefer concrete language and examples. Introduce technical vocabulary after the learner has seen the shape of the idea.