Skill UI

This skill provides professional, AI-powered code review, integrated with Codex AI. It is designed to ensure high code quality before commits, automatically generate comprehensive CHANGELOGs, and assist in managing large-scale refactoring projects efficiently.

Runs external LLM CLIs (OpenAI Codex or Google Gemini) to review uncommitted changes, branch diffs, or specific commits, providing an independent second opinion before merges or releases.

Vexor is a powerful command-line interface designed for advanced semantic search across local files. By utilizing vector embeddings and integrating with large language models (like Claude/Codex), it allows users to search files based on meaning and context, rather than just exact keywords. This is ideal for large codebases or document repositories where traditional keyword search methods are insufficient.

A comprehensive, multi-stage workflow for enhancing and structuring raw footage using advanced AI tools. It covers the entire pipeline—from initial recording (Screen Studio) through AI planning (Claude/Codex), deterministic cutting (FFmpeg), compositing (Remotion), asset generation (ElevenLabs/fal.ai), and final human polish (Descript/CapCut). Designed for users creating vlogs, tutorials, or polished corporate videos.

This skill implements an autonomous, multi-round review loop for refining complex research or codebases. It iteratively reviews content using specified external backends (e.g., Codex, manual review), implements necessary fixes based on feedback, and re-reviews until the positive assessment criteria are met or the maximum round limit is reached. It manages state persistence and allows for configurable difficulty levels, making it ideal for projects requiring rigorous, self-correcting quality assurance.

Get a deep, multi-round, and adversarial critical review of your research work, ideas, or drafts. This skill simulates the perspective of a senior ML reviewer (e.g., NeurIPS/ICML level) using dedicated external backends (Codex or Manual). It is designed to find logical gaps, missing experiments, and narrative weaknesses, providing brutal honesty to elevate your academic contribution.

A lightweight CLI tool for systematically comparing various coding AI agents (such as Claude Code, Aider, and Codex) head-to-head. It measures crucial metrics—pass rate, API cost, execution time, and consistency—on reproducible tasks defined via YAML manifests. Use this to objectively select the best agent for your codebase or to validate model updates before production deployment.

This CLI alias implements the PUA Loop protocol, designed for running structured, iterative verification cycles. It helps developers set up and manage complex coding tasks that require explicit verification commands and defined stop conditions. It serves as a powerful wrapper for AI coding assistants like Codex/Claude, ensuring robust and traceable development workflows.

This guide details how to use sophisticated coding agents (such as Codex, Claude Code, or Pi) through a structured bash workflow. It provides methods for complex development tasks, including building new features, refactoring large codebases, and reviewing pull requests (PRs). Key best practices covered include managing pseudo-terminals (PTY), defining focused working directories (`workdir`), and executing tasks in the background for robust, multi-step automation.

This tool audits the integrity of machine learning experiments by checking for common fraud patterns such as fake ground truth, score normalization fraud, and phantom results. It utilizes an external reviewer backend (like Codex or manual review) to perform an adversarial review across all project artifacts, ensuring reported results are trustworthy and methodologically sound.

This advanced skill is designed to generate high-quality, publication-ready academic illustrations and diagrams. It utilizes a sophisticated, multi-stage iterative workflow involving Claude for planning, layout optimization, and strict visual review, combined with a dedicated local Codex app-server bridge for native image rendering. The process ensures that generated figures meet the rigorous aesthetic and structural standards required for top-tier conferences and journals (e.g., CVPR, NeurIPS).

This skill provides comprehensive diagnosis and optimization for agent skill libraries. By analyzing historical session transcripts from platforms like Claude Code and Codex, alongside static skill file checks, it scores each skill across 8 dimensions. It identifies issues such as low trigger rates, poor post-invocation user reaction, and incomplete workflow execution, generating a prioritized report (P0/P1/P2) to significantly improve the agent's overall performance and reliability.