CodeCoach is a Chrome extension for coding-interview practice on LeetCode and Programmers. It adds progressive hints, debugging help, submission history, weakness tracking, and review scheduling directly beside the problem page.
The main design goal is to help with the next step without immediately giving away a complete solution.
Published on the Chrome Web Store: Install CodeCoach
Privacy policy: CodeCoach Privacy Policy
General-purpose coding assistants are useful for finishing code, but that is not always what I want while preparing for coding interviews. If the assistant reveals the whole approach too early, the practice session becomes much less useful.
CodeCoach is built around a different interaction: read the code I have already written, understand the current problem context, and give only as much help as I ask for. The extension then keeps track of the mistakes and topics that repeatedly show up so those problems can be reviewed later.
CodeCoach.mp4
The full walkthrough shows CodeCoach running on a real LeetCode problem, from contextual hints and debugging guidance through submission tracking and review features.
flowchart LR
PAGE[LeetCode / Programmers problem]
CONTEXT[Problem + editor + run result context]
PANEL[CodeCoach side panel]
AI[Progressive AI feedback]
LEARNING[Hints / weakness tags / snapshots]
REVIEW[History + review queue]
PAGE --> CONTEXT --> PANEL --> AI --> LEARNING --> REVIEW
The extension runs only on supported practice pages. It reads the visible problem context, current editor code, selected text, and visible run or submission result when those are needed for a feature.
Hints have three levels so the user controls how much of the approach is revealed:
- Light hint - a small nudge without naming the solution strategy
- Direction hint - points toward the relevant algorithm, data structure, or reasoning shift
- Specific hint - gives a concrete next step without returning a full accepted solution
The no-full-solution behavior is enforced in two places. Prompt instructions restrict complete answer code, and the side panel also checks model metadata and generated output before displaying hint-style responses.
CodeCoach can work from the code currently open in the practice editor and provide:
- full approach and complexity analysis
- selected-line explanations
- debugging help using visible errors or failing test output
- next-code-step suggestions
- before/after code comparison
- wrong-answer study-note generation
Visible pass/fail results can automatically create local code snapshots. Each snapshot records the current code, language, result state, elapsed practice time, and available run-result context.
Duplicate result events are filtered so the same submission is not repeatedly stored when page events fire more than once.
AI responses include structured learning metadata rather than only free-form text. CodeCoach records problem-type, caution-point, and implementation tags such as:
- binary search boundaries
- BFS / DFS traversal
- sliding window
- data-structure selection
- edge cases
- off-by-one mistakes
- time and space complexity
Repeated learning signals are aggregated into the History view so the user can see which topics keep causing trouble.
Problems can be added to a review schedule based on their result and study history. Review items can be completed, postponed, or removed, and failed attempts receive higher review priority than routine successful attempts.
A per-problem timer uses Chrome alarms and notifications so a practice session can be run under a fixed time limit without keeping the side panel open the entire time.
CodeCoach stores problem sessions, hint events, notes, code snapshots, review metadata, and learning events in Chrome extension storage. History can be searched and exported as Markdown for later review.
LeetCode and Programmers do not expose the same editor or page structure. CodeCoach uses site-specific content scripts plus several editor fallbacks to recover the current source code and problem context.
The current implementation can inspect Monaco, Ace, CodeMirror, and regular textarea/DOM fallbacks when available. It also normalizes platform-specific run/submission states so the rest of the extension can work with one internal context format.
AI requests use the OpenAI Responses API with streaming enabled. The background service worker reads the SSE stream and forwards incremental output to the side panel so hints appear as they are generated instead of waiting for the full response.
flowchart LR
EDITOR[Current editor code]
PROBLEM[Visible problem context]
RESULT[Run / submission result]
EDITOR --> REQUEST[Context builder]
PROBLEM --> REQUEST
RESULT --> REQUEST
REQUEST --> SW[Extension service worker]
SW --> OPENAI[OpenAI Responses API]
OPENAI -->|stream| SW
SW --> PANEL[Side panel]
The assistant returns hidden structured metadata alongside the visible response. CodeCoach validates and normalizes those tags before storing them, which lets the extension build weakness summaries and review history without trying to infer everything later from raw chat text.
Cloud sync is off by default. The OpenAI API key, guest-session state, saved notes, code snapshots, settings, and study history are stored in chrome.storage.local unless the user enables an optional sync feature.
Optional Firebase Authentication supports Google or email/password sign-in. When cloud sync is enabled, selected learning metadata such as hint events, weakness categories, review schedules, problem identifiers, and timestamps can be synchronized through Firestore. Saved code snapshots and wrong-answer note bodies remain local in the current release.
CodeCoach supports two AI access paths:
BYOK: Browser extension -> OpenAI API
Guest: Browser extension -> Firebase guestCoach -> OpenAI API
-> optional Firebase / Google services
BYOK requests use the user's own OpenAI API key and are sent from the extension directly to OpenAI over HTTPS. Guest requests use the CodeCoach Firebase backend to enforce free-use quota, cost controls, and abuse protection before forwarding the coaching request to OpenAI. OpenAI requests are made with store: false in the current implementation.
CodeCoach does not sell user data or use it for advertising. It does not run on arbitrary websites; host permissions are limited to supported coding-practice pages and the API endpoints required for OpenAI and optional Firebase features.
The full data-handling description is in the privacy policy, hosted from this repository.
Before publishing a build that enables guest AI, verify the Firebase project used by GUEST_ENDPOINT has:
- Firebase Anonymous Authentication enabled.
- Firestore available for guest quota and budget documents.
- The
OPENAI_API_KEYFirebase secret configured. - The
guestCoachfunction deployed at the endpoint insrc/background/coach-router.js.
The manifest and submission package target version 1.2.0. This repository does not
indicate whether Chrome Web Store review or publication has completed.
Before submitting 1.2.0, merge the release PR and verify the public privacy policy
shows version 1.2.0 and the date 2026-09-05. Keep the Web Store privacy disclosures
consistent with guest backend processing, optional authentication and sync, and
user-requested transmission of problem context, code attempts, and conversation history.
- Chrome Extension Manifest V3
- JavaScript, HTML, CSS
- Chrome Side Panel API
- Chrome Storage API
- Chrome Scripting API
- Chrome Alarms and Notifications APIs
- Chrome Identity API
- OpenAI Responses API
- Firebase Authentication
- Firebase Firestore for optional metadata sync
CodeCoach/
├── manifest.json
├── src/
│ ├── auth/ # sign-in UI
│ ├── background/ # service worker, AI streaming, timers
│ ├── content/ # LeetCode / Programmers page integration
│ ├── options/ # extension settings
│ ├── shared/ # prompts, storage, Firebase, taxonomy
│ └── sidepanel/ # main product UI and feature controllers
├── assets/
├── landing/ # React + TypeScript landing page source
├── index.html # English landing entry
├── ko/index.html # Korean landing entry
├── docs/ # generated GitHub Pages output and store docs
└── vendor/
Install the published extension from the Chrome Web Store.
git clone https://github.com/tkim602/CodeCoach.git
cd CodeCoachThen:
- Open
chrome://extensions. - Enable Developer mode.
- Select Load unpacked.
- Choose the repository folder.
- Open a supported LeetCode or Programmers problem.
- Click the CodeCoach extension icon to open the side panel.
An OpenAI API key can be added from the extension settings for BYOK AI features. Guest mode provides a limited free trial without requiring a user API key.
The public English and Korean pages share one React + TypeScript application. Vite builds both routes into docs/, which keeps the existing GitHub Pages URLs unchanged.
npm install
npm run devBefore committing landing-page changes, regenerate the Pages output and run the checks:
npm test
npm run typecheck
npm run buildCodeCoach is a practice tool, not an answer generator for active assessments. Its coaching prompts explicitly avoid complete accepted code by default and disable assistance for contests, assessments, certifications, hiring tests, private tests, and similar evaluation settings.
CodeCoach is an independent project and is not affiliated with LeetCode, Programmers, OpenAI, Google, or their parent companies.