Nothing to hand? Load the
— a CDN import, the browser cache switched off, a pipeline rebuilt inside the click
handler, a mistyped task id, no dtype, no dispose(), an uncapped
generation and a WebGPU request with nothing to fall back to — or the
,
where the correct verdict is ship-ready and the useful output is what to add next.
Paste the code — the prescan is free
No upload, no AI: the prescan reads your source in the browser and lists what it mechanically found. The pipeline, model, worker, env and tokenizer inventory, then the flags — a pipeline that is never dispose()d, pipeline() rebuilt inside a loop or a handler, a call with no dtype, no device chosen anywhere, device: "webgpu" with no navigator.gpu check, inference in a file that also drives the DOM, no progress_callback, an https:// CDN import, a model id with no revision, no try/catch around loading, an unrecognised task id, a call with no model id, generation with no max_new_tokens, the pipeline handle invoked once per item in a loop, useBrowserCache/useFSCache turned off, and literal Hugging Face tokens. Each group explains why it matters. This part costs nothing and happens while you type.
The AI reviews it — this is the metered part
A senior browser-ML engineer's pass: a shipping posture with the single most important change named, the inventory with each construct's role, and prioritized findings across performance, memory, correctness, loading UX, portability and hygiene — each with the problem, what the visitor actually experiences, the fix and a corrected JS/TS fragment. Every prescan flag is confirmed or explicitly set aside. Findings may only cite pipelines, model ids, options and files that actually appear in your code, and the reviewer is told not to state a model's size or latency as fact unless you supplied it. Pricing is honest: a worst-case amount is reserved before the run and only what the run actually uses is charged — the meter next to the button shows both.
Fix, export, re-run
Every corrected snippet in one paste-ready block, the findings as GitHub PR review comments with ```suggestion fences — paste each on the line it belongs to and GitHub can offer Apply — a tickable action checklist for the ticket, the findings table as CSV, and Markdown or JSON export of the whole review. Then start fixing: the prescan re-runs as you edit and the strip above the run button counts the flags you have cleared, the ones still open and any you have just introduced — in the browser, for free, before you pay for a second review. Run it again and the review itself is compared with the previous one: whether the posture moved, how the findings and high/critical counts changed, and which findings are gone, still reported or newly reported. Reviews are saved to your SkillSafe account when you are signed in, so they follow you to another machine; restore puts the code back in the form too, and says so plainly when the paste was too large to store in full.
Derived from the @huggingface/transformers-js skill (Apache-2.0 licence). Not affiliated with Hugging Face.