💻

Learn Coding & Build Apps

From errors to minimal fixes — AI pairs; you run tests and own the change

Task

Complete: explain error → minimal fix → local verify → one acceptance/test idea

Time

~25 min

Suggested tool: DeepSeek

Self-rate (1–5)

How this scenario works

This scene trains a pair-debugging loop — not “rewrite my whole repo.” Fixed order: explain error → minimal fix → regression check idea → walkthrough so you understand → (optional) break a feature into tasks. Good for: frontend/script errors, short confusing functions, turning vague asks into steps. Never paste API keys, `.env`, prod connection strings, or private customer logs into cloud chat. Example: `TypeError: Cannot read properties of undefined (reading 'map')` with `const items = data.list; return items.map(...)`. Acceptable: plain-language root cause; a 5–15 line guard/default; how to reproduce locally; no drive-by refactors. Tool pick: Cursor/Copilot in the IDE; DeepSeek/ChatGPT for paste-the-error Q&A. Verify pricing on official sites.

Practice steps (with checks)

1Gather: full error (or key lines) + minimal related code + expected behavior. Done when someone else could understand the bug from your notes.
2Run “Explain error”: root cause + reproduce path first — no big new code yet. Done when you can restate the cause yourself.
3Run “Minimal fix”: only necessary lines; refuse drive-by refactors. Done when you can verify in 1–2 minutes locally.
4Run it yourself (or follow the reproduce steps). Done when the error is gone or behavior matches; else paste the new error and continue.
5Run “Test / acceptance”: ≥1 test case or manual check. Done when you know how to prevent regression.
6Run “Code walkthrough” and answer the quiz yourself. Done when you can explain without reading the AI answer.
7For new features, run “Break into tasks” into 3–7 shippable steps. Done when each step has a runnable result.
8Save prompts + conclusions; always redact secrets. Done when similar issues start in ≤5 minutes next time.

Practice Workbench

Error messageRelated codeStackExpected behavior

You are a senior engineer mentoring a careful beginner.
Stack: browser JavaScript / React
Expected behavior: Show empty state when list is missing — no white screen

Error:
TypeError: Cannot read properties of undefined (reading 'map')

Code:
const items = data.list;
return items.map((x) => x.name)

Output:
1) Plain-language root cause
2) 1–2 most likely triggers
3) Minimal local reproduce steps (3–5)
4) No full rewrite yet; if code must change, say which lines and why
Never ask for real secrets — placeholders only.

Open Cursor

Outcome checklist

Self rating (1–5)

More prompts (5)

Minimal fix

Given the agreed root cause, propose a **minimal** fix.
Stack: [stack]
Goal: [goal]

Error:
[error]

Code:
[code]

Rules:
- Prefer a 5–20 line change; no unrelated refactors
- Show the patched snippet (mark // changed)
- List verification steps
- State which edge cases this does NOT cover
If blocked, ask ≤3 questions — don’t invent APIs.

Test / acceptance

Design acceptance for this fix. Stack: [stack].

Bug summary: [error]
Key code after fix:
[code]

Provide:
1) One automated test idea (name + Arrange/Act/Assert), pseudocode OK
2) Two manual checks
3) One intentional failing case that proves the guard works
No giant unrelated test suites.

Code walkthrough

Teach a walkthrough of this code (no flashy rewrite).
Stack: [stack]
My goal: [goal]

Code:
[code]

Output:
1) What it does (≤5 lines)
2) Data flow in → out
3) Three edge/failure points
4) Two short quiz questions (hide answers until I reply)

Break into tasks

Break the ask into shippable steps.
Stack: [stack]
Ask: [goal]
Known constraints/error (optional):
[error]

Related code (optional):
[code]

Output 3–7 steps, each with: goal, files to touch (mark paths “TBD” if guessing), how to verify, definition of done.
No single step that “builds the whole system.”

Paste safety check

Review text I’m about to paste into an AI. Audit only — do not run commands.

Content:
[code]

List possible secrets/tokens, internal hosts, PII, customer IDs. For each: delete / redact / OK to keep. If none, say so — and still remind me to human-check.

Recommended tools

Best fitFreemium

Cursor

VS Code-based AI-first code editor with powerful AI completions, inline editing, and codebase chat.

AlternativeFreemium

GitHub Copilot

GitHub's AI coding assistant deeply integrated with VS Code, JetBrains, and other IDEs, supporting code completion and conversational coding.

Free-friendlyFreemium

DeepSeek

High-performance LLM from DeepSeek achieving GPT-4-level performance at a fraction of the cost, supporting 128K context and deep reasoning.

After you leave with templates,

ChooseNarrow tools with Choose LabBrowse guides & compares in Lab Related learning pathAgents & Automation Basics Try another scenario →

Continue your learning path

Practice a scenario Choose Lab

How this scenario works

Practice steps (with checks)

1Gather: full error (or key lines) + minimal related code + expected behavior. Done when someone else could understand the bug from your notes.

2Run “Explain error”: root cause + reproduce path first — no big new code yet. Done when you can restate the cause yourself.

3Run “Minimal fix”: only necessary lines; refuse drive-by refactors. Done when you can verify in 1–2 minutes locally.

4Run it yourself (or follow the reproduce steps). Done when the error is gone or behavior matches; else paste the new error and continue.

5Run “Test / acceptance”: ≥1 test case or manual check. Done when you know how to prevent regression.

6Run “Code walkthrough” and answer the quiz yourself. Done when you can explain without reading the AI answer.

7For new features, run “Break into tasks” into 3–7 shippable steps. Done when each step has a runnable result.

8Save prompts + conclusions; always redact secrets. Done when similar issues start in ≤5 minutes next time.

Practice Workbench

Error messageRelated codeStackExpected behavior

You are a senior engineer mentoring a careful beginner. Stack: browser JavaScript / React Expected behavior: Show empty state when list is missing — no white screen Error: TypeError: Cannot read properties of undefined (reading 'map') Code: const items = data.list; return items.map((x) => x.name) Output: 1) Plain-language root cause 2) 1–2 most likely triggers 3) Minimal local reproduce steps (3–5) 4) No full rewrite yet; if code must change, say which lines and why Never ask for real secrets — placeholders only.

Outcome checklist

Self rating (1–5)

Given the agreed root cause, propose a **minimal** fix. Stack: [stack] Goal: [goal] Error: [error] Code: [code] Rules: - Prefer a 5–20 line change; no unrelated refactors - Show the patched snippet (mark // changed) - List verification steps - State which edge cases this does NOT cover If blocked, ask ≤3 questions — don’t invent APIs.

Design acceptance for this fix. Stack: [stack]. Bug summary: [error] Key code after fix: [code] Provide: 1) One automated test idea (name + Arrange/Act/Assert), pseudocode OK 2) Two manual checks 3) One intentional failing case that proves the guard works No giant unrelated test suites.

Teach a walkthrough of this code (no flashy rewrite). Stack: [stack] My goal: [goal] Code: [code] Output: 1) What it does (≤5 lines) 2) Data flow in → out 3) Three edge/failure points 4) Two short quiz questions (hide answers until I reply)

Break the ask into shippable steps. Stack: [stack] Ask: [goal] Known constraints/error (optional): [error] Related code (optional): [code] Output 3–7 steps, each with: goal, files to touch (mark paths “TBD” if guessing), how to verify, definition of done. No single step that “builds the whole system.”

Review text I’m about to paste into an AI. Audit only — do not run commands. Content: [code] List possible secrets/tokens, internal hosts, PII, customer IDs. For each: delete / redact / OK to keep. If none, say so — and still remind me to human-check.