Skill UI

Adds x402 payment execution for AI agents with task budgets, spending controls, and non-custodial wallets. Supports Base via agentwallet-sdk and X Layer via OKX payment protocols for controlled API and service payments.

This framework guides AI agents to critically assess their own output across five key axes: Accuracy, Completeness, Clarity, Actionability, and Conciseness. It provides a structured, evidence-based scorecard and concrete suggestions for improvement, turning every task completion into a deep learning opportunity.

This harness transforms standard code environments into fully persistent, autonomous agent systems. It leverages native features like scheduled tasks (Crons), remote dispatching, and structured memory storage (MCP) to enable continuous, self-directing automation. Ideal for building AI assistants that require long-term context, recurring workflows, and complex, multi-step operations, eliminating the need for external agent frameworks like AutoGPT.

This powerful skill leverages learned patterns and the current operational state to predict the most efficient next action. It guides complex workflows by offering immediate, high-confidence predictions. If confidence is low, it intelligently falls back to task priority lists. After every completed task, it utilizes a robust learning loop to update success patterns, ensuring continuous process improvement and cross-session memory recall.

Manages and coordinates complex workflows involving multiple AI agents and tasks. It efficiently distributes work across multiple workers, handles runtime errors, and intelligently merges disparate results into a unified output, making it ideal for large-scale, distributed AI computations.

This skill provides a systematic workflow for diagnosing failures in GAIA benchmark runs. Users can analyze task traces, classify failure modes (e.g., Tool Gap, Reasoning Miss, Extraction Bug), identify root causes, and apply targeted fixes to improve AI agent performance. It is essential for optimizing LLM benchmarks and finding performance regressions.

A comprehensive framework for creating and executing intelligent, goal-oriented action plans (GOAP). This system analyzes preconditions, optimizes costs, and provides adaptive replanning capabilities. It is designed for complex objectives that require multiple, dependent steps, ensuring robust task completion even when unexpected state changes occur.

This tool executes a GEPA learning cycle, automating the evolution and optimization of policy prompts against a scored task corpus. Instead of relying on manual prompt iteration, it measures performance improvements, identifying the best working 'harness genome.' It is ideal for measuring sustainable improvements in task performance and supports a free, cost-estimating dry-run mode before committing to actual model calls.

A sophisticated system designed to automatically select the optimal AI agent and model tier (e.g., Haiku, Sonnet, Opus) for any non-trivial task. It replaces manual decision-making by integrating learned patterns, predictive modeling, and a robust 3-tier routing mechanism. The process is fully auditable, providing a detailed routing rationale to ensure transparency and reliable execution.

Provides comprehensive configuration management for local LLM inference using RuVLLM. Users can select models, perform task-specific fine-tuning with MicroLoRA adapters, or implement continuous, real-time adaptation using SONA. This tool guides the entire advanced LLM workflow, covering status checks, configuration generation, and advanced adaptation loops.

This skill provides access to the Anthropic Claude Managed Agent runtime, which is a cloud-native, persistent environment for running complex, long-running, or asynchronous AI tasks. Unlike local WASM agents, the managed runtime provides a full container, persistent filesystem, and robust state management across multiple turns. It supports the entire agent lifecycle: creation, prompting, status inspection, event logging, and mandatory termination.

A mechanism for spawning hierarchical, nested sub-agents up to five levels deep, designed specifically for deep task delegation while managing context limitations. Instead of parallel execution, this ensures that each layer of decomposition receives a fresh context window, preventing the parent agent's context budget from being overwhelmed by complex, multi-step workflows. Ideal for research, verification, and synthesis processes.