Skip to main content

Agent harness and CLI to eliminate AI slop from frontend code

Project description

UIdetox

The anti-slop engine for AI-generated frontends.

UIdetox is an agent harness that eliminates the "AI smell" from your UI. It transforms generic, LLM-generated code into hand-crafted, high-fidelity interfaces through a systematic scan → fix loop that enforces design taste and detects common AI anti-patterns.

InstallationCommandsDesign Dials


From "Vibe Coding" to "Vibe Design"

AI coding tools are great at building things fast, but the results often scream "AI made this"—Inter font, purple gradients, glassmorphism, and generic hero dashboards.

UIdetox gives your agent an objective "Design Score" to optimize. It’s not a linter; it’s a standard. By knowing what not to do, your agent can finally produce code that makes people ask "Who made this?"—not "Which AI made this?"


The Agent Prompt

Paste this into your AI coding agent (Cursor, Windsurf, Claude, etc.) to begin:

I want you to eliminate "AI slop" from this codebase and make the UI indistinguishable from hand-crafted design. Your goal is to reach a **Design Score of 95+**.

To do this, use the UIdetox ecosystem (requires Python 3.11+). Run the following setup:
pip install uidetox
uidetox update-skill [your_agent_name]  # e.g., claude, cursor, windsurf, gemini
uidetox setup --auto-commit

THE LOOP
This is a closed-loop system. Do not stop after one fix. Repeat this cycle until the strict score is maximized:
1. Audit: Run `uidetox scan .` to queue anti-patterns.
2. Fix: Run `uidetox next`. This batches issues for the highest-priority file. Fix them holistically (including fullstack boundaries). 
3. Resolve: Run `uidetox resolve <ID> --note "What you changed"`.
4. Verify: Run `uidetox check --fix` to auto-commit formatting and catch/queue any new mechanical regressions incrementally.
5. Loop: Run `uidetox next` again.
6. Deep Verification: When the queue empties, run `uidetox rescan` to re-audit with fresh eyes. Keep looping until `uidetox status` shows Score >= 95.
7. Finish: Once the target score is reached, run `uidetox finish` to squash the aesthetic fixes cleanly onto your main branch!

RULES OF ENGAGEMENT
1. Heavy Lifting: Don't be lazy. Large refactors and small detailed fixes — do both with equal energy. Replace entire components if they reek of AI slop. Fix things properly, not minimally.
2. Subordinate Ego: `uidetox next` injects specific SKILL.md design rules into your context. Follow them strictly. Do not substitute your own analysis or rely on your baseline training data.
3. Orchestrator Mode: For complex passes, use `uidetox loop --orchestrator`. Act as a manager: generate prompts with `uidetox subagent`, delegate to parallel subagents, and use MCP (Model Context Protocol) servers to read the filesystem and execute terminal commands directly.

Installation

pip install uidetox

# Install the design rules for your specific environment:
uidetox update-skill claude    # (Also supports: cursor, gemini, windsurf, copilot)
uidetox setup --auto-commit    # Initializes design dials & git integration

Commands

The Engine (CLI)

Command Action
uidetox scan Full audit: auto-detects tooling and runs the 40-rule static slop analyzer.
uidetox next Batches the highest-priority issues with dial-calibrated design context injection.
uidetox status View your Design Score and actionable per-category hints.
uidetox loop Enter autonomous mode (creates a session branch and loops fixes).
uidetox rescan Clears queue and re-audits with 40+ fresh anti-slop rules.
uidetox finish Squash merges the autonomous session branch cleanly.

Design Skills (Slash Commands)

Use these for targeted improvements on specific files or directories:

  • uidetox polish – Final quality and alignment pass.
  • uidetox animate – Adds purposeful motion (spring physics, scroll reveals).
  • uidetox audit – Technical checks for accessibility and performance.
  • uidetox harden – Edge cases, error handling, and i18n.

Design Dials

Control the "aesthetic DNA" of the output by adjusting these values in uidetox setup:

  • DESIGN_VARIANCE (1-10): From clean/centered (1) to asymmetric/massive whitespace (10). Drives structural layout generation patterns (Bento grids vs split-screens).
  • MOTION_INTENSITY (1-10): From CSS-only (1) to complex spring physics and reveals (10).
  • VISUAL_DENSITY (1-10): From spacious "art gallery" (1) to "cockpit mode" data density (10).

The Slop Checklist

UIdetox actively hunts and destroys:

  • Use of Emojis in text content
  • Overused Inter font stacks.
  • Generic Purple-Blue gradients.
  • Identical Card Grids with no hierarchy.
  • Meaningless Glassmorphism.
  • Default Lucide/Radix patterns with zero customization.
  • "Generic Startup" copywriting (e.g. "Next-Gen", "Elevate").
  • Div Soup over semantic HTML.
  • Missing Focus/Hover accessibility states.

Credits & License

Built on the shoulders of desloppify and impeccable

MIT © OJamals

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

uidetox-1.2.0.tar.gz (229.4 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

uidetox-1.2.0-py3-none-any.whl (178.9 kB view details)

Uploaded Python 3

File details

Details for the file uidetox-1.2.0.tar.gz.

File metadata

  • Download URL: uidetox-1.2.0.tar.gz
  • Upload date:
  • Size: 229.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.13

File hashes

Hashes for uidetox-1.2.0.tar.gz
Algorithm Hash digest
SHA256 5b702d60c98d98863ce4b7387d5f95538a6876755c62a7a811fbc1066f6b31f8
MD5 a67184dd1afc051fbe998ea5cfe3d950
BLAKE2b-256 1339de24063102feeb0749bfdea25e26a70a0f49ab415c942c8199ebeefe5f6b

See more details on using hashes here.

File details

Details for the file uidetox-1.2.0-py3-none-any.whl.

File metadata

  • Download URL: uidetox-1.2.0-py3-none-any.whl
  • Upload date:
  • Size: 178.9 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.13

File hashes

Hashes for uidetox-1.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 168612c75e905ff32a0fc9bbfe6a0108de0e82feb97d445ea854d7730a4e7782
MD5 ebaf793ae207fec436b0be3828071001
BLAKE2b-256 5fd2aec3b498eb7e9f890f27227ded437595d1f51b19b23b9fc4fe0d73303e14

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page