HumanCursor

---

What HumanCursor actually does

The same idea for Selenium and Playwright: simulated cursor movement with variable speed and overshoot, aimed at the same behavioural analysis layer.

What HumanCursor's maintenance signals mean

pushed_at reads 2024-11-28 — 624 days, or about 1.7 years, before the field check, which places it inside the 365-to-1095-day band the rubric calls NRND.

NRND is borrowed intact from the electronic component catalogues, where it means Not Recommended for New Designs, and it is the most useful state in this register precisely because it is not a synonym for broken. The part is still in the catalogue. You can still buy it. If you are designing something new, do not design it in. Applied here: what this project does, it very likely still does. Starting a new build on it means inheriting a maintenance gap you did not create and cannot close.

No release was captured for this entry, which this register records as unknown rather than as none. The distinction matters more than it looks: a project with no tags and a project whose tags the sweep failed to read produce the same blank cell, and printing them as the same fact would be the register asserting something it has not measured.

Where HumanCursor sits in this catalogue

HumanCursor is filed in section 02, automation stealth plugins, patches & cdp-leak fixes, whose scope is this: Patches and plugins for Puppeteer, Playwright and Selenium, plus the proxies and cursor libraries that sit beside them. An entry sits in exactly one section here. Where a project could defensibly go in two, it is placed where a reader would look for it first rather than where it fits most tidily, and that is an editorial decision rather than a measured one.

Filed beside it in the same section: hazetunnel (NRND), ghost-cursor (ACTIVE), BrowserForge (SEMI). Comparing an entry against its own section-mates is more informative than comparing it against the catalogue as a whole, because everything in a section is answering roughly the same question and the differences left over are the ones that matter.

It is published under MIT, alongside the rest of the MIT entries in this register, and it sits with everything else currently marked NRND. Read those two facets together: a permissively licensed project that stopped shipping is not a safer dependency than a copyleft one that did not, and a filter on one axis alone will tell you the wrong thing.

Substitutes for the same job, with no claim that either is better: ghost-cursor. The alternative relation is symmetric and carries no ranking, because this register publishes no ranking of any kind. The full HumanCursor comparison table sets all of them side by side on licence, push date, release cadence and state.

What to settle before you adopt HumanCursor

Movement models age against the detectors that learn from them, so the interval since its last push is directly relevant to what it is for. Treat it as a starting point for your own path generation rather than a finished defence.