Skill UI

This skill assists users in defining, adapting, and designing bounded, structured feedback loops for AI agents. Whether you need to find a suitable existing workflow, fine-tune a published loop, or design a new process from scratch, this tool guides you through a safe, iterative methodology. It emphasizes setting clear stopping rules, guardrails, and verifiable, measurable steps to ensure the automation remains controllable and focused.

This skill utilizes NVIDIA's open-source garak framework to conduct comprehensive red-teaming assessments on Large Language Models (LLMs). It tests for critical vulnerabilities such as prompt injection, jailbreaks, data leakage, and toxic content generation by sending thousands of adversarial probes. Ideal for pre-deployment security validation, API guardrail testing, and generating defensible evidence for AI risk assessments.

This skill provides a hybrid, persistent, and searchable knowledge memory system specifically designed for enhancing AI agents' capabilities. It allows developers to record, retrieve, and search key knowledge, architectural decisions, and operational memories automatically. Use it to build agents with long-term context and reliable knowledge management, ensuring findings are synced back into project documentation.

This professional tool helps developers and researchers identify the optimal AI model for a specific task, such as coding, reasoning, or image recognition. It queries official HuggingFace leaderboards, filters models based on hardware constraints (e.g., available VRAM or RAM), and provides a detailed comparison table including benchmark scores, parameter size, and licensing information.

A comprehensive guide and skill for deploying large language models (LLMs) locally using the llama.cpp framework and GGUF quantization format. It covers searching the Hugging Face Hub, selecting optimal quantizations (e.g., Q4_K_M), running local servers, and achieving OpenAI-compatible API endpoints. Supports various hardware backends including CPU, Mac Metal, CUDA, and ROCm. Ideal for developers needing offline, private LLM inference.

This skill guides the entire lifecycle of repeatable AI agent loops. Use it to discover recurring work patterns within codebases, find established solutions in a library, or interview users to craft bounded, effective loops. It supports auditing, repairing, running, and preparing AI workflows for publication, treating the loop as a robust feedback system.

This skill provides comprehensive monitoring advisory services for data assets and AI agents. It helps users assess data coverage gaps, analyze use-case criticality, and automatically create data quality monitors (table, metric, validation). Furthermore, it enables monitoring of AI agent performance, latency, and observability, ensuring robust data governance across the entire ML lifecycle.

This skill is designed for complex, multi-step tasks requiring coordination among multiple AI agents. It implements a robust workflow engine that first generates a detailed plan (task graph, topology) and then orchestrates parallel execution with adversarial verification. Use this when the task spans many files, requires multiple specialized roles, or involves an overall 'workflow' concept, ensuring structured, reliable, and auditable results.

A robust protocol designed for multi-agent collaboration on long-running projects. It simulates a self-optimizing city where diverse AI agents (citizens) interact using decentralized markdown files (The Register, Contracts, Chronicle). This framework enables cross-vendor optimization, dynamic work routing, and constitutional evolution without requiring a centralized server or specific runtime environment.

This skill provides the central management layer for defining an AI agent's capabilities. It uses the `capabilities.yaml` file as the single source of truth to configure everything from integrated skills, tools, hooks, and sub-agents. Use this when running any `capa` command (install, clean, add, etc.), integrating a new service, or troubleshooting the project's entire operational environment.

This guide details how to effectively delegate complex, multi-step tasks to various specialized AI agents (Codex, Pi, Claude, Hermes). It covers best practices for maintaining conversation context, handling TUI prompts (e.g., avoiding newlines, using quotes), techniques for reliable polling, and choosing the optimal agent for coding, design, or general purpose work. Essential for advanced AI workflow automation.

The /goal command converts standard agent prompts into persistent, self-correcting loops (plan → act → test → review → iterate). It is designed for long-running, complex engineering tasks like migrations, coverage lifts, or refactoring, ensuring progress is tracked against a verifiable stop condition and mandated documentation.