v3.1.0: New skill "antislop-code", add per-skill READMEs, and others.

v3.1.0 ships the sixth skill: antislop-code. It cleans up code comments so generated code stops reading as obviously AI: remove the generic AI-slop comments, keep the valuable ones, and never touch the executable code.

Also in this release:

• Every skill folder now has its own README on GitHub, so each one explains itself to humans (devs and laypeople).
• New UI patterns: Filler Data in Fields and Columns, and Emoji as Decoration.
• Wizard step-1 fix: the agent never downloads the rules at runtime, you fetch them.
• Tagline updated to "Anti Slop: Rules for AI Coding Agents", so it covers all six skills, not just UI and copy.
• Roadmap spoiler: antislop as a plugin for other agents (Antigravity, Codex, Cursor, Gemini CLI, and others). Still a plan, no version promised, estimate Q3-Q4 2026.

v3.0.2: fix installer pointer, app UI patterns, adaptive python, copywriting patterns

Patch release. No rule changes, no renumbering. Closes two issues, merges four PRs.

Fixed

• #13: the installer wrote a Claude Code @import pointer into AGENTS.md, which Codex cannot read and Claude Code does not need.
• The CLI now writes a pointer block to the file each agent reads (CLAUDE.md for Claude Code, AGENTS.md for Codex and Antigravity), drops the 46 KB core import, and uses the same <!-- antislop:start/end --> markers as the wizard, so a second install replaces the block instead of duplicating it. A global install no longer creates agent folders for tools you do not have.

Added

• #14: antislop-ui had no patterns for app screens; all 28 were landing-page tells.
• New App & Dashboard section in antislop-ui: default dashboard shell, invented stat-card numbers, filler activity feed, charts without a question, generic table columns, placeholder empty and loading states. Six patterns, each citing rules that already exist.
• Two new copywriting patterns for sentences that hide their actor: Actorless Passive, and Inanimate Subject with a Human Verb.

Changed

• The contrast checker now runs on any platform: python3 on macOS and Linux, python on Windows, both pre-approved. No one is forced to install python3, and no permission prompt appears.
• The npm package author is now antislop.
• Version bumped to 3.0.2; install wizard URLs re-pinned

v3.0.1: Snyk fix, docs clarity, npm pkg fix, wizard re-pin

Patch release. No new features, no rule changes. Fixes the Snyk W012 warning and cleans up post-release feedback.

Fixed

• Snyk W012 (MEDIUM): the packaged core skill no longer instructs agents to curl skills at runtime. Step 4 of the packaged wizard is now "No download needed"; skills install through the picker (npx antislop-ai) or the plugin. allowed-tools no longer includes Bash(curl *). The manual antislop.md wizard is unchanged.
• npm publish warning: npm pkg fix normalized the repository field (string to object).

Changed

• Author split: the Claude Code plugin author and marketplace owner are miqdadbadjuber; the npm package (antislop-ai) author is antislop.
• Version strings bumped to 3.0.1 in the CLI (--version, banner) and the contrast server.
• Install wizard URLs re-pinned from tag v3.0.0 to v3.0.1.

Docs

• README: "Which agents" FAQ broken out per install path, Install section notes each agent's skill folder, new Contributing section with an open-issue link.
• guide.md: new "Which agents does it work with?" and "Feedback" sections.
• Badges flattened to a single height.

v3.0.0: as agent skills, plugin, and antislop-ai picker CLI

antislop is now installable. v3.0.0 packages the whole system as Agent Skills, a Claude Code plugin, and a picker CLI.

What's new:

• Five agent skills, one folder per skill (skills//SKILL.md), open Agent Skills standard
• The core (antislop) stays the lean, always-loaded filter: 38 rules, three tiers, Delivery Gate, Liveliness Toolkit
• Four additive skills: antislop-ui, antislop-copywriting, antislop-human (with the contrast checker), antislop-layoutmobile
• Two install doors from one repo: npx skills add miqdadbadjuber/anti-slop (skills.sh) and the .claude-plugin marketplace
• A picker CLI, npx antislop-ai: banner, skill list with the core locked on, project or global choice, writes the folders and the pointer
• The contrast checker as an MCP tool inside the plugin
• MIT license

Install:

npx antislop-ai

or straight from the skills directory:

npx skills add miqdadbadjuber/anti-slop --all

antislop v2.5.0: layoutmobile skill + 2 copywriting patterns

New

• antislop-layoutmobile, the fourth skill: a mobile layout must reflow into a distinct state, not stay a squeezed desktop. Covers breakpoints, scale, grids, overflow, tap targets.
• Mobile Navigation section inside it: Nav That Stays Desktop, The Bare Hamburger, Bottom Nav That Eats Content, Sticky Nav Steals the Screen. Skill checklist: 9 items.

Changed

• antislop-copywriting: two new patterns (Excessive Quotation Marks, All-Caps Emphasis) plus 2 checklist items.
• antislop-human: mobile layout mechanics now live in antislop-layoutmobile.
• Wizard URLs pinned to v2.5.0.

Housekeeping

• Generate-test passed twice: a subagent audit caught all planted mobile slop including every new navigation pattern, with correct R-03 citations and zero em dashes.

antislop v2.4.2

Two merged PRs from @fajarhide, plus housekeeping.

Fixes

• Skill checklists now say the polarity they actually use (PR #10, closes #9). The three skill checklists announced "any yes on a must-be-no item is a FAIL", but none of their 20 items is a must-be-no item. antislop-ui and antislop-human now say "All answers must be yes:", and antislop-copywriting says "Every line below must be true:". Before this fix, an agent reading the old header could invert every verdict and pass work the gate should stop.
• Docs cleanup (PR #8). ROADMAP marks v2.4.1 as shipped, and the manual-install section in README no longer uses an em dash.

Housekeeping

• The wizard URLs in antislop.md are re-pinned from v2.4.1 to v2.4.2, so fresh installs download the corrected skills.

v2.4.1: guide.md + first community fixes

What's new

• guide.md: a plain-English guide to antislop for people new to it. It explains what antislop is, who it is for, how to install it, what a skill is, and why one file is enough. No jargon.
• Wizard download URLs re-pinned to v2.4.1.

Fixes

Thanks to @fajarhide for finding these.

• #1: the R-35 gate item was the only one in Block 1 phrased as a positive, so an honest "no" meant "never ran the app". Reworded to state the defect like its 16 neighbours. R-03 added to the Part 1 Hard Gate list.
• #2: the contrast table said #777777 on white is 4.5 / borderline. It is 4.48 and fails. Every ratio now carries two decimals, and scripts/contrast-check.py --selftest recomputes all eight rows out of antislop-human.md so the table cannot drift from the formula again (verified under python -O).
• #3: README, ROADMAP and the pointer block still described the v2.3.0/v2.4.0 world. The pointer block now matches exactly what the wizard writes (marker line + three skills), the dangling antislop-layoutmobile load is gone, manual install fetches scripts/contrast-check.py (optional), version drift resolved, and the Indonesian .gitignore comment is now English.
• #6: antislop did not pass its own filter. Carve-outs written: em dashes in the R-XX/C-X headings, the Part 1 example, gate items and the Em Dashes teaching section; the **FORBIDDEN**/**REQUIRED** labels; and the - **Tell:**/**Why:**/**Fix:** skill headers. All are documentation structure, not UI text.
• #7: three gate items disagreed with the rule they cite. The R-37 gate now lets the labeled draft pass, the R-23 gate reads as AND, and the mobile-keyboard note cites R-03 instead of R-35.

Notes

• guide.md is a document, not a skill: it is not in the wizard menu or the skill table.
• Files: antislop.md, antislop-human.md, antislop-copywriting.md, scripts/contrast-check.py, guide.md, README.md, ROADMAP.md.

v2.4.0 · antislop-human

antislop v2.4.0 ships the third skill in the antislop system: antislop-human, a filter for the human side of a UI, so the result works for people with different eyes, hands, and setups, not just someone with a mouse and perfect vision.

What's new

antislop-human (new skill)

Four sections, each with the pattern, why it matters, and the fix (tell → why → fix), citing the core rule:

• Color & Contrast: low-contrast text, text over a photo or gradient, the grey-on-grey hallucination, non-text contrast, and The Contrast Checker
• Keyboard: removed focus outline, mouse-only patterns, broken tab order
• Focus & States: weak or invisible focus indicator, color-only feedback, missing UI states
• Zoom & Mobile Use: text that cannot zoom (WCAG 1.4.4), mobile keyboard covering the form

The Contrast Checker (new)

The checker is antislop's answer to agents that claim WCAG AA without checking. Three layers, so it never blocks on any one of them:

• scripts/contrast-check.py: run it for an exact ratio and a PASS/FAIL verdict for normal (4.5:1) and large (3:1) text
• The WCAG 2.x formula: written out, for agents that cannot run code
• A reference table: 8 common pairings with verified ratios

The skill makes the agent prove contrast, not assert it. #555 on black is 2.82:1 and fails; an agent that eyeballs it says it passes.

Mobile layout is its own skill

The human skill keeps the parts about people (zoom, keyboard covering the form). The layout mechanics behind mobile (breakpoints, grids, overflow, tap targets) move to a new skill, antislop-layoutmobile (v2.5.0), so the roadmap is now six skills.

Install wizard update

The wizard in antislop.md now offers four options (All / antislop-ui / antislop-copywriting / antislop-human) with a short "which concern do I need" description. Choosing human also downloads the contrast-check script into scripts/. Skill downloads are pinned to this release tag.

How to update

Existing users: add the new skill, the script, and one pointer line.

curl -o antislop-human.md https://raw.githubusercontent.com/miqdadbadjuber/anti-slop/v2.4.0/antislop-human.md
mkdir -p scripts
curl -o scripts/contrast-check.py https://raw.githubusercontent.com/miqdadbadjuber/anti-slop/v2.4.0/scripts/contrast-check.py

Then in your entry file's antislop block, add:

- People: antislop-human.md

New users: download antislop.md once, tell your agent to read it, and the wizard walks you through the rest.

Files in this release

- antislop.md (core + wizard)
- antislop-ui.md (skill, unchanged)
- antislop-copywriting.md (skill, unchanged)
- antislop-human.md (new skill)
- scripts/contrast-check.py (new script)

What's next

antislop-layoutmobile (v2.5.0), antislop-docs (v2.6.0), antislop-identity (v2.7.0), then v3.0.0: the whole system packaged as an installable skill/plugin. See ROADMAP.md.

v2.3.0 · antislop-copywriting

antislop v2.3.0 ships the second skill in the antislop system: antislop-copywriting, a copy and text filter that catches AI-slop writing the same way antislop-ui catches slop UI.

What's new

antislop-copywriting (new skill)

26 AI-writing patterns in four sections, each with a before/after example and its core rule cited:

• Tone & Voice (8): empty AI vocabulary, significance inflation, empty claims and social proof, weasel attributions, persuasive authority tropes, chatbot closers, fake-candid openers, signposting
• Rhythm & Structure (7): rule of three overuse, negative parallelism, aphorism formulas, staccato drama, synonym cycling, false ranges
• Honesty & Evidence (3): fabricated specifics, speculative gap-filling, generic positive conclusions
• Hygiene & Markdown (6): em dashes, boldface overuse, inline-header lists, emojis in headings, filler phrases, excessive hedging

Plus the parts that keep it a filter, not a ban list:

• What NOT to flag: a false-positive guard. Look for clusters of tells, not isolated ones; don't gut real human writing.
• Signs of human writing (preserve these): stops the filter from over-editing.
• Voice calibration (optional): when the user provides a writing sample, match it instead of scrubbing to a sterile default (R-37).
• No fabrication: a rewrite never adds facts, names, numbers, dates, or citation36, R-38).
• Draft, audit, final: a delivery process with two audit questions.
• Copywriting Skill Checklist: a copy-specific supplement that runs with the cor

Install wizard update

The wizard in antislop.md now offers three options (All / antislop-ui / `antisloshort "which skill do I need" description. Skill downloads are pinned to this releasetag.

README redesign

README gets a table of contents, a 30-second intro with before/after examples, a "Which skill do I need?" section, a cleaner skill table, and an FAQ.

How to update

Existing users: add the new skill and one pointer line.

curl -o antislop-copywriting.md https://raw.githubusercontent.com/miqdadbadjuber/anting.md

Then in your entry file's antislop block, add:

- Copy & text: `antislop-copywriting.md`

New users: download antislop.md once, tell your agent to read it, and the wizard wal

Files in this release

- antislop.md (core + wizard)
- antislop-ui.md (skill, unchanged)
- antislop-copywriting.md (new skill)

What's next

antislop-a11y (v2.4.0), antislop-docs (v2.5.0), antislop-identity (v2.6.0), then v3.le skill/plugin. See ROADMAP.md(https://github.com/miqdadbadjuber/anti-slop/blob/main/ROADMAP.md).

v2.2.0: Core + First-Run Install Wizard + antislop-ui skill

What's new in v2.2.0

antislop is now a system: a lean, always-loaded core (antislop.md) plus optional additive skills, one per concern, shipping one per version.

First-Run Install Wizard (in the core)

• New install flow: download antislop.md once, tell your agent to read it, and the agent walks you through choosing skills, downloading them, and setting up the pointer for you.
• An offer, not a requirement: antislop.md alone remains a complete filter.
• The wizard asks which skills to install (multi-select, "All" recommended), downloads the chosen files into the same folder as the core, appends a managed pointer block at the end of your entry file (AGENTS.md / CLAUDE.md / GEMINI.md / etc.) without touching existing content, and asks the usage-mode question.
• Honest direction handling: if no DESIGN.md exists, antislop recommends having one and offers three paths (user-supplied direction, agent-supplied with a warning, or skip with the "draft without direction" label).

New skill: antislop-ui (UI / visual)

• Deep-dives the UI/visual concern: color, layout, components, decoration, structural flow, and motion. 28 patterns in the tell-why-fix shape, each citing the governing core rule (R-XX), plus a UI Skill Checklist that runs alongside the core Delivery Gate.
• Additive, never duplicated: references core rules by number and loads with the core, never instead of it.

Docs

• README rewritten as the system index with the new install instructions.
• ROADMAP updated: v2.2.0 marked done; antislop-copywriting is next (v2.3.0).

Backward compatible

• Existing single-file users keep a complete filter.
• Existing manual pointers are merged, not duplicated.