Post
CommunityJev Media & Talks
How EmDash uses Clef to moderate the plugin registry
EmDash CMS screens every plugin listing in its registry with Cloudflare's Clef decision model: nine typed questions on the listing text and links, eight on each icon and screenshot, and a probability threshold of 0.45 on any category. The registry itself is built on AT Protocol.
- Format
- Post
- Reading time
- 3 min
- Published by
- Community
- Author
- Scott Buscemi
- Last verified
- Oct 2026
Highlights
- Every listing is screened by Clef before it appears on plugins.emdashcms.com; the AT Protocol records underneath stay with their authors.
- Nine questions cover the text and links, including sexual content, hate, graphic violence, phishing, impersonation, scams, spam and deceptive links.
- Each icon and screenshot gets eight corresponding questions, while links are judged from their text and destination rather than from screenshots.
- Clef returns a probability per category, and the post says a result of 0.45 or higher sends the listing onward; the fetched text cuts off at that sentence.
Watch out
Copy reviewed against the fetched blog text on 2026-10-15; the fetch was truncated at about 12,000 characters in the middle of the sentence after the 0.45 threshold, so what happens to a flagged listing and any appeal or cost details could not be read. Everything here is EmDash's own description of its workflow, not an independent test of Clef's moderation quality.
Questions about this entry
Short answers drawn from the linked project or page.
Every plugin listing in the official EmDash registry is screened by Cloudflare's Clef decision model before it can appear on plugins.emdashcms.com: the labeler asks Clef nine questions about the text and links, and eight for each icon and screenshot.
Because Clef answers bounded questions with typed probabilities that application code can act on rather than generating prose, and it accepts images. EmDash says those two properties fit a plugin listing that mixes structured metadata and links with visual assets.
Text and links cover explicit sexual content, hateful or dehumanizing content and graphic violence; phishing or credential requests, impersonation, scams and spam; and deceptive links, misleading media or claims, and attempts to manipulate moderation. Icons and screenshots get corresponding questions.
The post says links are evaluated from their text and destination rather than from screenshots, which avoids flagging ordinary screenshots of spam filters and link-checking tools as deceptive or abusive.
Clef returns a probability for each category, and the post says a result of 0.45 or higher sends the listing onwards. The fetched text cuts off mid-sentence at that point, so the exact consequence - review, hold or rejection - is not visible in what was read for this entry.
The registry is built on AT Protocol, so plugin authors publish signed releases from their own accounts. EmDash does not control the underlying records, but it decides what appears in its catalog, which the post frames as keeping that open publishing model from becoming an open door for abuse.
Scott Buscemi wrote it for the EmDash CMS blog on 7 October 2026 as a three-minute read. It describes EmDash's own workflow, so it is a vendor account rather than an independent evaluation of Clef's moderation quality.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
YouTube: Building a contextual brand-safety scorer on Jev from scratch.
X: A website slop detector that checks 35 tells in 243 ms
X: Breaking, golden nugget, or slop
X: A filter for reply-guy comments
More Jev resources like this
Moderation for communities and apps, powered by Jev: every message gets a probability for spam, scam, harassment, nsfw, doxxing and more, plus rules in plain English. Thresholds and actions are yours, and flag-only is the default.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
ohernandezdev/jevmodPython
ohernandezdev/jevmodPythonAlex dc's Spanish-language explainer on Jev for the Agentes IA newsletter: what the TypeSafe decision model is, what it is not, and how to try it, opening with the headline claims of 193x faster and 444x cheaper.
Jev Media & TalksPost#community#catalog-import#madewithjev
Community
Cloudflare's announcement of Clef-omni, an open-weight decision model that takes audio and video alongside text and images in one request, built on a Qwen3-Omni mixture-of-experts backbone, plus a faster Clef and a cheaper Clef-flash.
Jev Media & TalksPost#community#catalog-import#madewithjev
Community
Microsoft's announcement of Microsoft-Decision-1, a model for fast decision scoring built for routing, classification, prioritization, verification and workflow control, available in Microsoft Foundry and through OpenRouter. Published 2026.10.09.
Jev Media & TalksPost#community#catalog-import#madewithjev
Community
Stephen Solka's Hugging Face article tests whether Jev can filter political fundraising from an inbox: a frozen 100-email eval, one noul question per email, then follow-on experiments that shrink the classifier to CPU-only inference with SetFit and ONNX Runtime.
Jev Media & TalksPost#community#catalog-import#yibie
Community
An intent compiler for AI agents: it probes what files and tickets already answer, asks only what they can't, and emits a typed IntentSpec YAML saved to .intent/<intent>.intent.yaml before work is routed. The README positions it as the input layer for routers and typed-decision models like Jev and Laya.
Jev Agents & MCP#community#catalog-import#madewithjev
Community
angel291592/intent-routerPython
angel291592/intent-routerPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.