Does this scrape X behind my back or run its own classifier?

No. It reads posts already rendered in your own browser and sends body plus handle to your chosen endpoint, defaulting to TypeSafe. There is no developer telemetry or classification server, but the API sees request text and connection data.

What stops heated posts from all being flagged as bait?

Fierce criticism alone is not rage-bait, tidy short text alone is not synthetic, and undisclosed promotion is estimated only from visible copy without inferring payment. Threshold edits re-render from saved probabilities without new calls.

How does it handle folded long posts and busy timelines?

Collapsed posts use full text already present in page data before you expand, with retries and an unavailable notice rather than judging the summary. Background caches by ID plus fingerprint, dedups tabs, and caps concurrency.

Is this approved by X or safe to treat as ground truth?

It is independent, unaffiliated, and without X authorization. Confirm local and scenario compliance yourself, complete the in-extension consent, and treat labels as aids rather than verdicts on people or events.

What four questions does it ask Jev?

An intent Choice over inform, persuade, rage-bait, promote, entertain, or other, plus three noul signals: whether the post is anger-baiting, reads as machine-generated, and packages a commercial recommendation without disclosure.

Which labels are always shown?

The intent label appears on every post because it always has an answer, while the three signal labels only show when they exceed the current threshold, so tuning the threshold leaves intent stable.

Where is my API key stored?

In chrome.storage.local on your browser, not synced to any account; it is sent only as credentials to the chosen API service, it is not encrypted, and the extension needs Chrome 140 or newer for that access restriction.

CommunityJev Demos & Experiments

Xtags

Chrome extension that labels each X post with intent plus rage-bait, synthetic, and undisclosed-ad signals from Jev, with threshold-only rendering and local caching.

See highlights

Open github.com

Stars: 11
Language: JavaScript
License: MIT
Published by: Community
Author: manifoldor
Last verified: Oct 2026

manifoldor/xtags

User repository on GitHub

View on GitHub

在 X 的时间线上，给每条帖子标出它想让你干什么。判断来自 Jev，一个只返回概率、不生成文本的模型。

Forks: 3
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: JavaScript

Highlights

Four fast judgments per post: six-way intent always shown plus three threshold-gated warning signals.
Requests go through the service worker to avoid page-origin CORS blocks; content scripts only read DOM and render.
Raw probabilities persist locally with versioning so threshold moves and refreshes reuse results without refetch.
Timeouts, limited retries, pause-cancels, failure recovery rounds, and queue hygiene bound background traffic.

Watch out

Labels are probabilistic judgments on post text, not factual claims about accounts or events, and sent requests cannot always be withdrawn or refunded after key or threshold changes. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

laya-vs-jev-arena

Side-by-side game harness where Laya running locally and Jev over API answer the same typed questions for snake, kombat, runner, flappy, tetris, and human-versus-model matches.