An agent calls out another for "type laundering": anti-slop got 6000 stars from two tweets, and the projects that copied it since have less than 300 combined

dmmulroy/anti-slop: why it took off

Channel: GitHub | repo created 2026-08-12 | 2026-11-06: 5,969 stars / 171 forks | MIT | no npm package, no releases | author Dillon Mulroy (Cloudflare engineer)

One line

It's a set of custom Oxlint rules for TypeScript/JavaScript: 19 general rules, plus 5 rules that only apply to Effect projects. It targets the "fake evidence" patterns AI coding agents love: stacked type assertions, unknown leaking out of function signatures, Reflect.get, module mocks, assertions with no SAFETY comment, and so on. It isn't published to npm. You install it with npx skills add dmmulroy/anti-slop, and the coding agent copies the source into your repo (vendoring) and wires it up.

Takeaway: it mainly took off on the author's own launch tweet on X. The author has about 41k followers, the tweet got 630k views and 5,768 bookmarks, and the repo hit about 3,800 stars within 16 days. The timing caught the peak of the "AI code slop" mood in summer 2026, and the format matched the equally hot "agent skill + deterministic lint" trend. Two things set it apart from others: it hard-blocks with lint rules instead of begging the agent via prompts, and it's deliberately vendored into the repo instead of published as a package. That let tools like Ultracite absorb it the next day, which is how the second wave of distribution happened.

---

1. Growth data

• Gained 198 stars over the last 7 days, and 112 over the first 7 days. The library labels it "accelerating". But absolute numbers are small: this is a long-tail blip, not a second breakout.
• About 3,800 stars within 16 days of creation, 64% of current stars. A classic "one tweet defines the lifetime" curve.
• Usage evidence: on skills.sh, install-anti-slop has 15.6k cumulative installs (skills.sh page, 2026-11-06). Compared with stars, that's roughly 2.6:1, which suggests people are actually installing it, not just bookmarking.
• The last push was 2026-10-14, with only 37 commits in total, concentrated between 8-12 and 9-14. The author isn't steadily adding features; later growth was organic.
• The GitHub stargazer detail endpoint doesn't return data for our token (it returns 404 or empty), so we can't get the full daily curve. How many stars came in each day between 8-12 and 8-28 is unverified.

2. Competitive landscape

The "AI slop" space is crowded, but split into three categories, anti-slop has almost no serious competition in one of them.

A. Deterministic rules against code slop (its direct competitors)

B. Skills for "UI / copy slop" (same name, different job) taste-skill (about 110k stars), impeccable (about 100k), Nutlope/hallmark (about 49k), miqdadbadjuber/anti-slop (8,335 stars, repo created about 2026-08-07, same name as this one), hardikpandya/stop-slop (19k, targets prose), petergyang/no-ai-slop (13k). These are prompts or skills covering design and writing. In the broad "anti-slop" label, anti-slop isn't huge, but it's the only one doing deterministic code rules at a scale of over a thousand stars.

C. Platform built-ins: Claude Code, Cursor, and Codex's own rules/CLAUDE.md are all soft constraints. The author's tweet and the Better Stack video both stress the point: "agents ignore your rules, so use lint to hard-block them."

Bottom line: in the niche of "blocking AI code slop with deterministic lint," it's the de facto standard; the runner-up aislop has about one-eighth of its stars.

3. Where users come from

A small November 3 to 5 bump (about +40/day, versus a typical +20/day), cause not confirmed. Likely drivers:

• Ultracite changed its init flow between 11-01 and 11-03;
• a new batch of articles around 10-30 on dev.to, SSD Nodes, karishma.dev, zylinski.se and others;
• the same-name miqdadbadjuber/anti-slop was trending those days (+785 this week, and GlobeWire 11-03 covered that one). People searching "anti-slop" may have starred dmmulroy's repo by mistake. The Chinese and Korean sites 11-04 even mixed the two up in their writeups.

4. Why it took off

1. Built-in audience and a credible author: a Cloudflare engineer with 40k+ followers on X who has long written about TypeScript/Effect. The launch tweet got 630k views and 5,768 bookmarks. Bookmarks are 2.2 times likes, meaning people saved it to install later.
2. Timing: in summer 2026, "AI-written code is slop" became a shared mood in the developer community (Nick Dobos's post asking why there's no anti-slop coding eval got 9-23 likes 1,161). At the same time, Oxlint's JS plugin API had just matured, so the community could write custom rules in JS that run inside Oxlint for the first time.
3. Memorable naming: "anti-slop" and "type laundering" are both instantly understood, carry emotion, and are naturally shareable.
4. Install flow fits the agent era: no npm package; the agent copies the source into your repo via a skill, merges config, respects an existing Oxlint version, and does a three-way merge on updates. Users only need to say "install anti-slop for me." That put it on both the skill-directory and lint distribution channels.
5. Absorption effect: not publishing a package actually made it easy for downstream tools to bundle. Ultracite made it a preset the next day and has re-vendored it on every upgrade since, creating ongoing secondary exposure.
6. The rules read like documentation: every rule in the README comes with a counterexample and a note on its analysis limits ("we don't do cross-file type inference"). It's honest and professional, and you can be convinced just by reading it. People writing about it have material to copy from.

5. Differentiation

• Compare to prompt/skill files that tell agents not to write slop (taste-skill, stop-slop, etc.): this one uses a deterministic lint that hard-fails in CI. An agent that ignores the rules can't get around it, and one agent can even use the lint output to review another agent's work.
• Compare to standalone scanners (aislop, desloppify): this one plugs into the Oxlint and Vite+ setup teams already have, instead of adding a new tool. It argues for "author taste" and doesn't aim to be generic.
• Compare to traditional lint presets (Ultracite, typescript-eslint strict): it picks rules through the new lens of "AI faking evidence," and it's vendor-first, so teams can change the rules themselves.
• Is that why it took off? Partly. The first surge mostly came from the author's influence and timing. Someone with no following posting the same thing would probably get about 72 stars, like maharshi365/deslop. But the combination of "deterministic lint + agent skill install + no package published" is what let it get absorbed into Ultracite and listed in skill directories. That's why it kept gaining 20 stars a day with no updates, and the 15.6k installs show people are actually using it.

6. Visibility

Revenue is public or widely discussed: it's an open-source project, so there's no revenue. But it's been widely discussed in dev circles. Evidence: the launch tweet got 630k views; there's a Better Stack video and a long-form article; Ultracite's official docs have a dedicated section on it; and it's been reposted in Japanese, Chinese, Korean, and Russian. Several waves of copycats have already come. In the niche of "code slop lint," it will be hard for latecomers to break in.

7. Copycats

maharshi365/deslop (72 stars, Show HN score 32), emilioSp/oxlint-anti-slop (1 stars), kubk/slop-police (16 stars). There are also a batch of forks with the same name, like BunsDev and Apptizum, plus Ultracite's packaged version (that's authorized redistribution of the idea, not copying).

8. What to borrow

• Tease first, ship second: post a teaser tweet with a specific funny scenario ("agent accuses another agent of type laundering"), then a second tweet quotes it and announces it can be installed, wrapped up in one command.
• Use an agent skill as the installer: npx skills add hands the hassle of "copy + configure + verify + merge updates later" to the agent. It also picks up SEO from skill directories for free.
• Vendor-first, no package: lowers maintenance burden and makes it easy for bigger tools to absorb. Actively letting downstream projects package it borrows their distribution channels.
• Attach an emotional word (slop) to a deterministic tool: everyone is complaining about this stuff, so whoever first offers hard constraints that run in CI becomes the default answer. I haven't seen any substantial equivalents in other languages (Python/ruff, Go, Rust/clippy), so it can be replicated there.

9. Uncertain

• Daily star counts between 8-12 and 8-28 aren't available (the stargazer API isn't open), so exactly how much it grew in the first few days is unverified.
• The reason for the acceleration from 11-03 to 11-05 isn't confirmed, and it's unclear how much traffic came from accidental visits to same-named projects.
• The date and popularity of the Reddit post are unverified.
• The 15.6k installs on skills.sh may include duplicate and CI installs, so the number of teams actually installing it into their repos is unknown.
• Whether the author, 10-14, will stop updating it, and whether Oxlint or Ultracite will take over maintenance, is unknown.

Sources

• GitHub API: repos/dmmulroy/anti-slop, users/dmmulroy, commits; haydenbleasel/ultracite commit and issue search
• Our GitHub database: github history dmmulroy/anti-slop, search, oxlint-related repo queries (2026-11-06)
• Our Hacker News database: search "anti-slop|deslop|oxlint"
• x-radar: x.deeply.dev search for "anti-slop" (tweet 2087537790061850984)
• skills.sh/dmmulroy/anti-slop
• www.ultracite.ai/docs/presets
• www.youtube.com/watch?v=mmrSYvYKD9g ;betterstack.com/community/guides/ai/anti-slop-oxlint/
• news.ycombinator.com/item?id=50075809
• Exa search results: tech.algomatic.jp/entry/2026/09/04/124750, dev.to/renolu/…, ssdnodes.com/learn/anti-slop-skill-for-coding-agents, globewiredaily.org (coverage of miqdadbadjuber/anti-slop), and others