The world's trailing AI Lab - Made with ❤️ in Berlin

A comprehensive ecosystem of tools designed to enhance the AI-assisted development experience, improve transparency, and ensure proper attribution in AI-human collaborative coding.

🍦 Brain Freezes — get-togethers for the SLICC crowd. Upcoming & past events →

AI CLI tool wrapper with worktree support. Launches AI coding assistants (Claude, Codex, Cursor, Gemini, etc.) with appropriate bypass flags and optional isolated git worktrees for safe experimentation. Supports multi-agent mode with parallel execution and automatic cleanup.

Git wrapper that enforces safe practices for AI coding agents. Prevents dangerous operations like git add ., ensures proper AI attribution, enforces commit hooks, and maintains human sign-off on all AI-generated commits.

Transparent GitHub CLI wrapper that automatically detects AI tool usage and ensures proper bot attribution. Uses device flow to exchange user tokens for bot tokens, so actions appear as "as-a-bot[bot] on behalf of @user" rather than appearing to come directly from the user.

Transparent Xcode toolchain shim that shadows xcodebuild, xcrun, swift and xcodegen on PATH, so there is nothing to teach the agent — no CLAUDE.md rule, no skill file, no MCP server. Leases simulators by UDID so parallel agents stop colliding on one device, isolates DerivedData per worktree, refuses long runs the caller has not budgeted for, and retries simulator infrastructure failures without retrying genuine test failures. Humans get an untouched pass-through.

GitHub App token broker running on Cloudflare Workers. Provides user-to-server GitHub tokens via device flow for ai-aligned-gh, ensuring actions show proper user attribution with app badges rather than appearing as bot-only actions.

Claude Code skill for pushing status updates and action items to TRMNL e-ink displays via webhook API. Includes responsive templates for full screen (message + action items list), half screen (top/bottom split), and smaller layouts (message only), plus a comprehensive TRMNL Framework design system reference.

A browser-based coding agent that runs as a Chrome extension, CLI server, or Electron float. Runs Claude directly in the browser with full filesystem access, a WebAssembly shell, browser automation via CDP, and a complete suite of code editing tools. The self-licking ice cream cone - an AI coding agent that builds itself.

GitHub CLI extension that intelligently filters and highlights errors in GitHub Actions workflow logs by severity. Auto-detects failed PR checks, prioritizes errors (fatal → error → warn → fail), and shows context around matches. Perfect for developers and AI agents working with limited context windows.

Quickly install Claude/Agent skills from other repositories. Works standalone or as a GitHub CLI extension. Copies skills, creates discovery scripts, and updates AGENTS.md with clear markers for idempotent updates.

A curated catalog of 30+ ready-to-install agent skills for SLICC and other AI coding agents — browser automation, email & chat (Gmail, Outlook, Slack, Teams), music (Suno, Strudel, Apple Music), documents (PDF, PPTX, presentations), research, and screenwriting. Each skill is a self-contained directory (SKILL.md plus scripts, references, and assets). Install them with upskill ai-ecoverse/skills --all.

Ranked GitHub triage as a gh extension. Shows what needs your attention now: PRs/issues waiting on you, work waiting on others, local repos behind, AI coding sessions to resume, categorized notifications, and cleanup candidates. "Garfield was wrong. Mondays are when everything gets sorted." Install with: gh extension install ai-ecoverse/gh-monday.

GitHub CLI extension that hunts down stale git worktrees across your machine — showing each one's last-touched age and disk size, oldest first — and reaps the ones you've forgotten. Read-only by default (--reap to delete), safety-classifies worktrees as clean/dirty/unpushed/orphan, and steers clear of macOS privacy-protected folders so it never trips a permission prompt. The perfect companion for the worktrees YOLO and other agents leave behind. "Seasons don't fear the reaper." Install with: gh extension install ai-ecoverse/gh-reaper.

GitHub Action that analyzes repository git history to determine what percentage of commits were made by AI tools. Generates dynamic badges showing AI contribution levels with smart logo selection based on the dominant AI tool (Claude, Cursor, Gemini, Copilot, etc.).

A lightweight MCP secrets manager that uses Cloudflare Workers for compute and GitHub for persistence. Stores secrets as GitHub Actions Secrets and retrieves them via triggered workflows with ephemeral key encryption. Exposes five MCP tools over OAuth 2.1 with dynamic client registration, working with any MCP-compatible client.

Lightweight shell library for detecting AI coding agents. Provides robust two-phase detection (environment variables + process tree) to identify when code is running under AI control. Powers the AI detection in ai-aligned-git and ai-aligned-gh.

Kev decision models running in the browser on WebGPU. Ask yes/no, multiple-choice and rating questions about a piece of text and get calibrated probabilities back — no server, nothing leaves the tab. Speaks the TypeSafe System One API, ships Kev-0.8B, 4B and 9B as int8 bundles on Hugging Face that match the full-precision models' accuracy, and has a live demo. The -vision bundles also take an image, read by Qwen3.5's own vision tower; in decision-vision-bench Kev-4B vision is the best browser model for image decisions. Install with npm install @ai-ecoverse/kev.js onnxruntime-web.

Cua's cua-s1 form-filling decision model in the browser. It reads each field of an HTML form and a document's Label: value pairs and decides whether to fill the field with one of those values, tick it, click it, or leave it alone. A 3 MB ONNX graph plans a whole form in about 150 ms on the CPU, matching Cua's PyTorch model to 3.2e-6. Nothing is typed until you apply the plan. Weights on Hugging Face, live demo (including PDF input). Also runs Cua's cua-s1-4b-0.2 on WebGPU, which picks the next action on a screen from its accessibility tree or a screenshot (demo). Install with npm install @ai-ecoverse/cua-s1.js onnxruntime-web.

Jev-Omni, a Gemma 4 12B decision classifier, in the browser on WebGPU. Give it a state, a question and up to 256 options, optionally with an image, and get one probability per option from a single forward pass. The 13.6 GB int8 bundle gives the same answers as the full-precision model on all 293 DecisionBench questions (86%). It is the strongest of these models on long text decisions. Work in progress: video and audio are still to come.

Kev.js, cua-s1.js and jev-omni.js side by side on one mixed set of 411 image and text decision questions, each model in its own request format, in Chrome on WebGPU. Reports accuracy with confidence intervals, calibration, latency, download size and memory, plus a plain recommendation of which model suits which job.

• Safe Experimentation: Use YOLO with worktree mode to test AI-generated changes in isolation
• Proper Attribution: Ensure all AI contributions are clearly marked in git history and GitHub
• Efficient Debugging: Quickly find critical errors in CI/CD logs with gh-workflow-peek
• Skill Sharing: Share and install AI agent skills across projects with gh-upskill
• Transparency: Track and visualize AI contributions with vibe-coded-badge-action
• Disk Hygiene: Find and reap forgotten git worktrees by age and size with gh-reaper
• Shared Simulators: Stop parallel agents colliding on one iOS Simulator with ai-aligned-xcode
• In-Browser Decisions: Route, classify and score text on the user's own GPU with kev.js
• Form Filling: Plan which document values go into which form fields, in the browser, with cua-s1.js
• Image Decisions: Answer yes/no, choice and rating questions about screenshots and photos in the browser with kev.js vision, compared in decision-vision-bench

Each tool can be installed independently. Visit the individual repository links above for detailed installation instructions and usage examples.

Most tools offer one-line installation:

# YOLO
curl -fsSL https://raw.githubusercontent.com/ai-ecoverse/yolo/main/install.sh | sh

# ai-aligned-git
curl -fsSL https://raw.githubusercontent.com/ai-ecoverse/ai-aligned-git/main/install.sh | sh

# ai-aligned-gh
curl -fsSL https://raw.githubusercontent.com/ai-ecoverse/ai-aligned-gh/main/install.sh | sh

# ai-aligned-xcode (macOS only)
curl -fsSL https://raw.githubusercontent.com/ai-ecoverse/ai-aligned-xcode/main/install.sh | sh

# gh-upskill
curl -fsSL https://raw.githubusercontent.com/ai-ecoverse/gh-upskill/main/install.sh | bash

# gh extensions
gh extension install ai-ecoverse/gh-workflow-peek
gh extension install ai-ecoverse/gh-upskill
gh extension install ai-ecoverse/gh-reaper

# am-i-ai (detection library, used by ai-aligned-git/gh)
curl -fsSL https://raw.githubusercontent.com/ai-ecoverse/am-i-ai/main/install.sh | sh

# kev.js (decision models in the browser)
npm install @ai-ecoverse/kev.js onnxruntime-web

# cua-s1.js (form-filling decisions in the browser)
npm install @ai-ecoverse/cua-s1.js onnxruntime-web

These tools embrace vibe coding - a new AI-assisted software development paradigm where developers guide AI tools through natural language to build software. The ecosystem ensures:

1. Transparency: All AI contributions are clearly attributed
2. Safety: Dangerous operations are prevented or clearly flagged
3. Accountability: Human developers maintain oversight and sign-off
4. Efficiency: Tools reduce friction in AI-human collaboration
5. Visibility: Analytics show the real impact of AI on codebases

All projects are open source under Apache 2.0 or MIT licenses. See individual repositories for details.

Contributions are welcome across all projects! Please visit individual repositories to submit issues or pull requests.

We are on the Hub as @ai-ecoverse — models, datasets and Spaces as they land. So far: ai-ecoverse/kev.js, the browser-ready Kev weights behind kev.js, and ai-ecoverse/cua-s1.js, the ONNX export of Cua's form-filling model behind cua-s1.js.