antislop is a rule set for removing “AI slop” from the output of AI coding agents. AI slop refers to the generic output AI tends to mass-produce: templated UI designs you feel you have seen before, copy full of hype, and code comments that explain far too much. It is an open-source project released in August 2026 by Miqdad Badjuber, an independent developer based in Indonesia, and distributed on GitHub under the MIT license. You load it as skills into the agents its README lists as supported, such as Claude Code, Codex and Cursor.
What sets it apart is that it acts purely as a “filter” and never dictates colors, fonts or layout. The design direction itself is left to a design spec the user provides (such as a DESIGN.md), and antislop’s only job is to subtract the “AI feel” from it.
Key Features
- 38 rules in three tiers: Rules R-01 through R-38 are grouped into three tiers: “Hard Gate” (absolutely banned), “Purpose-Gate” (allowed when there is a reason) and “Quality Locks” (for consistency)
- Three metrics to tune the result: So that filtered output does not end up bland and lifeless, three dials (ENERGY, RHYTHM and MOTION) adjust how lively the screen feels, its rhythm and the amount of motion
- PASS/FAIL check before delivery: A verification gate (Delivery Gate) requires a PASS/FAIL-style check report to run before any output is delivered, making it easier to catch rules that were missed
- Skills split by area: On top of the core antislop skill, there are add-on skills for UI (antislop-ui), copywriting (antislop-copywriting), accessibility (antislop-human), responsive design (antislop-layoutmobile) and code comments (antislop-code), so you install only what you need
- Multiple ways to install: Choose from the
npx antislop-aiinstaller,npx skills add miqdadbadjuber/anti-slopvia skills.sh, plugins for each agent, or a manual install that pastes a single file (antislop.md) into the chat
Pricing
| Plan | Price | Key Features |
|---|---|---|
| Open source (MIT) | Free | All 6 skills available. Modification and redistribution allowed |
Pricing is as of September 2026. There is no mention of paid plans or enterprise licenses. For the latest information, check the official repository.
Usage fees for the agents themselves (Claude Code, Cursor and so on) are charged separately.
Pros & Cons
✅ Pros
- It rarely clashes with your existing brand guidelines or design spec, so you can reduce the “AI feel” while keeping your current design rules intact
- You can install only the area you are struggling with (just UI, just copy, just code comments), which keeps the amount of instructions the agent has to read small
- You can adopt it without changing the coding agent you already use, so there is no cost of switching tools
- The pre-delivery check reduces the burden of manually running through a checklist every time
- It is free, and you can read the rules and rewrite them to suit your own work
⚠️ Cons
- It does not provide a direction such as color scheme or tone, so you need to prepare your own design spec to get good results
- It is a new project released in August 2026 and is still being updated every few days. There is no auto-update, so you have to update manually for each install method
- The rules are written in English, and there is no confirmed track record of how well they work on Japanese copy or Japanese UI
- As an open-source project by an individual developer, there is no enterprise support
Comparison with Similar Services
Several projects share the “anti-slop” or “antislop” name, and each targets something different.
| Service | Difference |
|---|---|
| antislop (this article) | Agent skills that cover UI, copy and code comments across the board. A filter that does not dictate design direction |
| skew202/antislop | A linter that detects AI slop in code through static analysis. UI and copy are out of scope |
| dmmulroy/anti-slop | Oxlint-based and focused on code quality checks for TypeScript/JavaScript |
| dharmawan-id/anti-ai-slop | A rule set focused on English and Indonesian prose. Also offered as an MCP server and CLI |
| sam-paech/antislop-sampler | A technique that suppresses slop vocabulary through token sampling at inference time. A different domain from agent skills |
Who Is It For
- People who feel that every screen their AI coding agent builds ends up looking the same
- People who want to tone down exaggerated phrasing in announcements or blog posts drafted by AI
- Developers who want to clean up the over-explained comments attached to AI-generated code
- Teams that have their own design spec but find agent output drifting away from it
Summary
Against the problem of AI output that looks plausible yet familiar, antislop takes the position of removing common patterns without imposing a direction of its own. Because it works best in combination with your own design spec, an easy way to start is to install a single skill for the area you are struggling with, such as UI or copy, and see how the output changes. The project is updated frequently, so it is worth checking the release notes from time to time.