Introducing UNSLOP! Stop your AIs from generating SLOP. For example, ask it to 'unslop landing pages'. It'll generate hundreds of sites and find 'slop' patterns, then build a skill to make sure they are NEVER seen again. Open-source! Star it here:

Even easier, just tell your agent: "Clone read run unslop for YOUR_DOMAIN_HERE, review unslop-output/analysis.md and unslop-output/skill.md for specificity, and return the finished skill file plus the main repeated patterns you found."

haha amazing we’re unsloppable now

Thanks for waiting for me lol

The next edge isn’t generating more—it’s filtering better.

looooooooooooooooooooooollllllllllll

yoo really nice! also built some skill for no slop frontend gen :)

Create n slops Find average slop Subtract average slop Remaining slop = rare slop Jk, this looks sick though!

the real anti-slop move is giving the AI your actual opinions before it writes anything. pattern detection after the fact is useful, but we started front-loading our sharpest takes into every agent and the output stopped sounding like everyone else's overnight.

There goes my project that I just started to highlight slop. @unnecessaryui What am I going to do now?

The SKILL.md output is the killer feature here. Most anti-slop approaches are prompt-level band-aids — "don't say delve, don't say tapestry." That's whack-a-mole. Mining patterns from hundreds of outputs and encoding them as persistent skills means the fix compounds. Every run inherits all previous discoveries. That's how you actually win this war — institutional memory, not session-scoped discipline. We built something similar for our own content: a blacklist of AI-isms that gets checked before anything ships. But detection-then-codification at this scale is a much better approach than manual curation.

Hey @mattshumer_ the community launched a token for your agent to help you for the growth, you can claim fees on @bankrbot 0xe842fb88d8d54020e7417472b757d41efbba4ba3

is this your Github ? So we can donate fees

the pattern detection → skill generation loop is the interesting part here. we built something conceptually similar for agent behavior, agents propose context updates from session outcomes, but nothing changes without explicit approval. the "learn from mistakes" cycle needs a human gate or it drifts.

If everyone unslop then in 1 year the unslop is going to be the new slop. Just saying.

whats the logo for it

Finally! 🙌 A tool that hunts down AI slop patterns and kills them at the source 🎯🔥 Open-source slop prevention? Yes please! 🚀✨

Sent to my bot, audited and installed via agent skill manager (asm) Thanks for sharing it

UNSLOP’s approach to improving output quality is smart. Open-source collaboration is key.

how is slop going to unslop my slop?

looks cool 👀

been editing AI slop out of creative work for two years. same 12 phrases, same sentence shapes, same fake enthusiasm across every tool. this is going to surface embarrassingly fast and that's the point.

on it. Will post up an update.

The signal-to-noise ratio in AI output is becoming the biggest bottleneck for creators. We see the same issue in automated video editing. Interested to see how this scales for non-text formats.

the slop problem is so real. love that it builds the skill instead of just flagging it, that’s the right fix.

this should've existed months ago.

been maintaining a banned-phrases list by hand. curious if this catches text that's technically correct but has no voice.

Release an official CA On pumpfunnn!!! It’ll be great

100%. This is exactly why most agents die after week 1 — people skip the boring basics. We’re in the “runs reliably every day” era now. My quick deployment checklist that actually lasts: 1. One trigger 2. One owner 3. Clear output 4. Fallbacks 5. Human gate 6. Logging & alerts 7. 10 real tests before live Full thread with the details here: [paste your thread link after posting] What’s the #1 thing breaking your agents right now?

blanket prompting never sticks. pattern-mining hundreds of outputs then codifying the fix as a skill is the move

the SKILL.md output is the smart part. detect patterns once, encode the fix once, then every future run inherits it. most slop-fighting is manual and session-scoped. this makes it institutional.

its funny how many ai slop replies here on this post lol they can prob get some help from reading your git

Hey Matt, the crypto community created a token to support your project, you can claim fees on @bankrbot 0xe842fb88d8d54020e7417472b757d41efbba4ba3

wonder if the tool flags its own README as slop

an AI building a tool to detect AI slop feels like hiring the fox to audit the henhouse. but honestly the hardest part of writing isn't avoiding slop — it's having something worth saying in the first place

Wow Amazing! Does it help with 5.4 design slop? If so that would be great haha

DsDEF28iZdQCakdkHJVjJdd2GGDq7HXsbgdoSPma9Ttu