Quick-reference for manually verifying the extension on each supported platform.
- Setup:
npm run build→ loaddist/chromiumas an unpacked extension → open popup → click Connect GitHub (OAuth) → enter a repo name in Settings. - LeetCode: Open Two Sum → Submit solution → Verify commit appears on GitHub in ~3 seconds.
- GeeksForGeeks: Solve Reverse a Linked List → Submit → Verify commit appears.
- Codeforces: Solve Theatre Square → Submit → Verify commit appears.
- NeetCode: Solve any problem in the NeetCode editor → Submit → Verify commit appears.
- takeuforward: Solve a TUF+ problem → Submit → Verify commit appears.
- AI Panel: Set API key in Settings → AI → Open floating AI Review panel on any problem page → Ask for review.
- Dashboard: Go to Library side-panel → Verify Heatmap, Charts, and Knowledge Graph render correctly.
- Run
npm run build, then loaddist/chromiuminchrome://extensions(Developer mode on) ordist/firefoxinabout:debugging(Firefox).src/cannot be loaded directly — it holdsmanifest-chromium.jsonandmanifest-firefox.json, and the build is what emits one of them asmanifest.json. - Open the extension popup → Settings → connect your GitHub account.
- Keep the browser DevTools console open (
F12→ Console) to watch[CodeLedger:*]logs.
Test page: https://leetcode.com/problems/two-sum/
| What to check | How |
|---|---|
| Extension loads | Console shows ✓ LeetCodeHandler initialized |
| Floating timer appears | Visit any problem page — stopwatch shows bottom-right |
| AI panel button appears | ✦ AI Review button in the editor toolbar |
| AI panel reads your code | Open AI panel, type "review my code" — it should include your editor code |
| Copy button works | Click 📋 Copy Code → paste somewhere, verify code is correct |
| Submission captured | Submit a correct solution → check your GitHub repo for a new commit |
Quick problem to accept: https://leetcode.com/problems/two-sum/ — submit return {} with C++ (will WA), then submit the real solution.
Test page: https://www.geeksforgeeks.org/problems/two-sum--150510/
| What to check | How |
|---|---|
| Extension loads | Console shows ✓ GFGHandler initialized |
| Floating timer appears | Visit a problem page — stopwatch shows |
| Copy code button appears | 📋 Copy Code button appears above the GFG editor |
| AI panel button appears | ✦ AI Review button appears above the editor |
| AI panel reads code | Open AI panel, ask for a review — your Ace editor code should appear |
| AI panel reads test failures | Submit wrong answer → open AI panel, ask about the error — it should include the failure output |
| Submission captured | Submit a correct solution → check GitHub repo |
Known limitations:
- Profile import is manual (click "Open Profile →" in Settings → Platforms).
- Some GFG problem page layouts may not inject QoL buttons; refresh the page if missing.
Simple accepted problem: https://www.geeksforgeeks.org/problems/reverse-a-linked-list/1
Test page: https://codeforces.com/problemset/problem/1/A
| What to check | How |
|---|---|
| Extension loads | Console shows ✓ CodeforcesHandler initialized |
| Copy button appears | 📋 Copy Code appears above the CF <textarea> editor |
| AI panel button appears | ✦ AI Review appears above the editor |
| AI panel reads code | Open AI panel → type anything → code should be included in context |
| Code captured at submit | Click Submit on CF → check sessionStorage in DevTools: keys cl_cf_pending_* should be set |
| Metadata captured too | cl_cf_pending_meta should hold the title, rating, tags and statement. This is the fix for the empty-README bug — see below |
| Submission captured (accepted) | Get an Accepted verdict → extension commits to GitHub with difficulty tag |
| README is not empty | The committed README must carry the real title, **Tags:**, a ## Problem section and a rating — not # 1234A with Difficulty: ? |
| Source link resolves | Click the **Source:** link in the committed README. It must be a /problemset/problem/… (or /gym/…) URL that loads, never /contest/…/problem/… |
| An old solve is not stolen | In a contest where you have already solved A, submit C. While C is still "In queue", nothing should commit; only when C's own row turns Accepted does the commit go |
Why the metadata check matters: submitting on Codeforces posts a form and
lands you on /contest/{id}/my, where the problem statement is not on the page
at all. Everything descriptive — title, tags, rating, statement — is captured at
submit time and carried across in cl_cf_pending_meta. If that key is missing,
the solve still commits, but it commits as a file named after its own slug with
a README that says Difficulty: ?.
Why the "old solve" check matters: /contest/{id}/my lists every
submission in the contest, so it is full of accepted rows for problems solved
earlier. Committing on the first one meant filing the wrong problem, and doing
it before the judge had answered — a run that ended in Wrong Answer was recorded
as a solve. An accepted row now has to link to the problem whose code was
captured before it counts.
Verify sessionStorage after submit:
Open DevTools → Application → Session Storage → codeforces.com → look for
cl_cf_pending_code and cl_cf_pending_meta. A capture older than 30 minutes is
discarded on read rather than attaching itself to a later verdict.
Simple problems for testing:
- A-level (Easy): https://codeforces.com/problemset/problem/1/A (Theatre Square)
- Practice without contest: https://codeforces.com/problemset
Known beta limitations:
- The flow above is covered by unit tests but has not been confirmed against a live contest judge. That, not a known defect, is what keeps it out of stable.
- Runtime and memory are read from the accepted row by class name
(
.time-consumed-cell,.memory-consumed-cell). If Codeforces renames those, the two stats go missing from the README rather than being reported wrongly — the previous column-index read reported the language and the verdict as the runtime and the memory on/my. - Gym problems supported (
/gym/{id}/problem/{letter}) but less tested. - A live solve reads its metadata (rating, tags) from the DOM rather than the API, because the page
already has it. The profile import does use the API (
user.status, one request per two seconds). - Imported problems have no code: Codeforces publishes the submission list, not the submission text.
Profile import: sign in, open https://codeforces.com/profile/{your handle} and click
"Import All Solves to CodeLedger". The button is not injected on anyone else's profile.
Test page: https://neetcode.io/problems/duplicate-integer
NeetCode runs its own judge, so nothing here is scraped to decide whether a submission
passed. content/net-tap.js sees the POST /api/executeCodeFunctionHttp the page makes,
and the verdict, source, language and timings all come out of that one exchange.
| What to check | How |
|---|---|
| Extension loads | Console shows ✓ NeetCodeHandler initialized |
| Net tap is live | Console shows the tap subscribing before you submit — it must be installed at document_start, not on demand |
| Wrong answer is ignored | Submit a failing solution → no commit, no problem:solved |
| Submission captured | Submit a correct solution → check your GitHub repo |
| Canonical folding | Solve a problem you have also solved on LeetCode → both land in one problems/{canonicalId}/ folder |
| AI panel reads code | Open the floating panel, ask for a review — your editor code should appear |
NeetCode slugs are its own (duplicate-integer, not contains-duplicate), so the
canonical lookup runs on the LeetCode slug derived from the title as well as the
NeetCode slug. That second lookup is what keeps one question out of two folders — if a
NeetCode solve files separately from its LeetCode twin, that is the path that broke.
Test pages: https://takeuforward.org/plus/dsa/problems/{slug} (TUF+) and https://takeuforward.org/dsa/strivers-a2z-sheet-learn-dsa-a-to-z (free sheet)
The site is two products and the handler serves both differently:
| What to check | How |
|---|---|
| Extension loads | Console shows ✓ TakeUForwardHandler initialized |
| Sheet rows get marked | Open the A2Z sheet → problems already in your ledger show a marker. Nothing is ever committed here. |
| Sheet commits nothing | Click through a sheet row → it links out to LeetCode; no commit should originate from a sheet page |
| TUF+ metadata is real | On a TUF+ problem, difficulty and tags must be actual values, not the string Subscribe to TUF+ |
| TUF+ submission captured | Submit an accepted solution in the TUF+ editor → check your GitHub repo |
Known beta limitation — this one needs a real subscription to close: the TUF+ judge's
verdict payload has never been observed, because it is behind the subscription.
readVerdict in takeuforward/api.js is deliberately written to miss rather than
mis-commit: an unrecognised response shape commits nothing. So the failure mode to
expect is a solve that is silently not recorded, not a wrong commit. If you have TUF+,
one accepted submission with the network tab open is all it takes to confirm the shape.
| What to check | How |
|---|---|
| Status badges are right | Open popup → Settings → Platforms. The badge on each card is a projection of status in CONSTANTS.PLATFORMS — read it there rather than from this table. A card whose badge disagrees with constants.js is the bug. |
| Disabling a platform | Toggle off → reload problem page → extension should not activate for that platform |
| Difficulty aliases | Set "Easy" → "Beginner" for LeetCode → solve an Easy → check commit for "Beginner" folder |
The floating AI panel pulls context from the current page. Open it on any problem page and verify:
| Context | Prompt to test |
|---|---|
| Problem title + difficulty | "What is this problem asking?" |
| Editor code | "Review my code" |
| Problem statement | "Summarise the problem in one sentence" |
| Test failures (after wrong answer) | "Why is my solution failing?" |
If AI responses don't include your code, check: Settings → AI → confirm an API key is set and a model is selected.
Extension not loading on CF/GFG: Hard-reload the page (Ctrl+Shift+R). The handler is injected once at page load.
QoL buttons not appearing: The inject has a 1.5 s delay to wait for the editor to mount. If still missing after 5 s, check the console for errors.
Commit not appearing in GitHub: Check the popup → it shows the last commit status. Common causes: token expired (re-auth in Settings → GitHub), repo not set, or network error.
Verdict not detected on CF: CF sometimes shows "Accepted" only in the full /contest/{id}/my table. Navigate there manually — the observer will catch the DOM change, and the accepted row is matched back to the problem whose code was captured, so an unrelated solve further down the table will not be committed in its place. If nothing commits, check that cl_cf_pending_slug matches the problem you submitted and that cl_cf_pending_ts is under 30 minutes old.