Releases·Tool Release
unflat: one CSS block that gives flat AI-built sites a ground, a light and a few mounted surfaces, in their own colors
A developer released 'unflat', an Agent Skill for Claude Code, Cursor and Codex that fixes the flat 'AI mockup' look of agent-built sites. The agent audits the page, extracts design tokens, and appends one reversible CSS block adding a ground color, a light source, and mounted surfaces using the site's own colors. No markup changes are required.
3 engagement1 sourceSun, Sep 6, 2026, 08:40 PM
AI Overview
- unflat is an Agent Skill for Claude Code, Cursor and Codex (v1.2, one week old) that audits a page and appends one reversible CSS block — no markup changes.
- The block adds a darker ground, one light source, mounted sections with the site's own corner radius, an accent-colored wash, and one motif, keeping hero and strips on the ground.
- Install via `npx skills add merturl4576/unflat`; deleting the block reverts the page.
- The author says every example was produced by the skill itself on a mid-tier model, and the repo logs what broke and what was fixed.
Entities
Related
Products·launched
NPCTalk: AI agents give game NPCs voiced, in-character speech via Fish Audio
NPCTalk is a modding toolkit that lets players talk to NPCs in any game and get spoken, in-character answers about a second later, using Fish Audio cloned voices. LLM coding agents (Claude Code, Codex, Cursor, etc.) wire it up via skills, a self-hosted voice server, and tools. It solves the problem of static, scripted game dialogue for modders and players, demonstrated with a full GTA V example.
NPCTalk is a modding toolkit that lets players talk to NPCs in any game and get spoken, in-character answers about a second later, using Fish Audio cloned voices. LLM coding agents (Claude Code, Codex, Cursor, etc.) wire it up via skills, a self-hosted voice server, and tools. It solves the problem of static, scripted game dialogue for modders and players, demonstrated with a full GTA V example.
49 engagement1 sourcegithubThu, Oct 15, 2026, 11:37 PM
Products·launched
Open-source Claude Code skill that makes AI-generated sites look less like AI slop
A Claude Code skill that guides the model through a structured design process when building web pages, avoiding common AI design tells like gradient headlines, fake terminal windows, and mismatched components. It works by having Claude pick one design direction, plan visual elements, and pull fitting components. It targets developers who use Claude Code to build sites but find the results generic.
A Claude Code skill that guides the model through a structured design process when building web pages, avoiding common AI design tells like gradient headlines, fake terminal windows, and mismatched components. It works by having Claude pick one design direction, plan visual elements, and pull fitting components. It targets developers who use Claude Code to build sites but find the results generic.
8 engagement1 sourceredditThu, Sep 17, 2026, 09:06 AM
Products·building
Team chat where AI coding agents are members of the channel
Hilos is a Discord-style team chat that integrates AI coding agents as channel members. It solves the problem of isolation when using coding agents like Claude Code, Codex, or Cursor by allowing the team and agents to communicate in the same space, sharing context and feedback without manual copy-pasting.
Hilos is a Discord-style team chat that integrates AI coding agents as channel members. It solves the problem of isolation when using coding agents like Claude Code, Codex, or Cursor by allowing the team and agents to communicate in the same space, sharing context and feedback without manual copy-pasting.
1 engagement1 sourceredditTue, Jul 21, 2026, 05:25 PM
Products·launched
Oqoqo: Build realistic evals and custom benchmarks for agent tasks
Oqoqo is a platform for creating realistic evals and custom benchmarks for real-world tasks, measuring how agent-friendly product surfaces are against various AI coding agents. It helps developers regression test MCP, CLI, skills, SDK, and other agent-facing interfaces, and allows creating and sharing custom benchmarks.
Oqoqo is a platform for creating realistic evals and custom benchmarks for real-world tasks, measuring how agent-friendly product surfaces are against various AI coding agents. It helps developers regression test MCP, CLI, skills, SDK, and other agent-facing interfaces, and allows creating and sharing custom benchmarks.
1 engagement1 sourcehackernewsMon, Aug 10, 2026, 09:27 PM
Products·launched
CurseBench: benchmark for swearing at coding agents
CurseBench is a local, read-only tool that scans conversations from coding agents like Claude Code, Codex, Cursor, OpenCode, Antigravity, and Grok to compute a score based on how often users swear at the model. It uses regex over chat stores with zero runtime dependencies, solving the problem of quantifying frustration in AI-assisted coding. Built for developers who want to measure their own behavior or benchmark agent interactions.
CurseBench is a local, read-only tool that scans conversations from coding agents like Claude Code, Codex, Cursor, OpenCode, Antigravity, and Grok to compute a score based on how often users swear at the model. It uses regex over chat stores with zero runtime dependencies, solving the problem of quantifying frustration in AI-assisted coding. Built for developers who want to measure their own behavior or benchmark agent interactions.
8 engagement1 sourceredditTue, Aug 11, 2026, 04:40 AM