Skill UI

An internal, non-user-invocable agent used by the meta-skill creator process. It executes positive (G3) and negative (G4) smoke tests against a candidate meta-skill definition (SKILL.md). It simulates meta-resolution using LLM-generated fixtures and reports the results in a structured JSON format.

SkillOpt-Sleep simulates an agent's nightly 'sleep' cycle to continuously improve performance without requiring model retraining. It harvests historical sessions, mines recurring tasks, and replays them offline. Learning and insights are then consolidated into the agent's memory and skills (CLAUDE.md/SKILL.md), proposing bounded, validated edits that the user must explicitly adopt, ensuring measurable, sustained improvement over time.

A generic entry point for delegating self-contained, multi-step tasks to specialized LLM agents (Codex, Claude Code, Pi). Utilizes background processing tools to manage the full lifecycle of AI-driven sub-tasks, such as policy review, complex report synthesis, document generation, or large-scale code refactoring. Ideal for meta-skill workflows requiring full agent agency.

A comprehensive minimal terminal coding harness designed for building, extending, and integrating advanced AI capabilities directly into CLI environments. It supports deep integration via SDK (Node/TypeScript), RPC, and JSON event streams. Developers can create custom skills, providers, TUI components, and utilize advanced features like web search, video understanding, and agent orchestration.

This skill allows users to mine their extensive coding history across various AI agents (Claude, Codex, Gemini, etc.) into a persistent, structured corpus. It surfaces reproducible workflows as evidence and interactively helps the user forge these past practices into reusable, installed skills across their agent ecosystem. It is designed to synthesize judgment from raw transcripts into actionable, portable skills.

This skill provides a comprehensive guide for building scalable Retrieval-Augmented Generation (RAG) pipelines and persistent agent memory using Pinecone as the vector database. It covers index setup, embedding large corpuses, implementing dense and hybrid semantic search, and advanced metadata filtering for real-world production systems. Use this for building searchable knowledge bases and context-aware AI agents.

This Main Agent acts as a central orchestrator for a specialized squad of AI agents, managing complex software development lifecycles. It guides projects through defined phases—from requirements gathering and planning to architecture, implementation, testing, and deployment. It maintains strict context discipline by summarizing agent outputs and referencing full artifacts, ensuring efficient, structured, and non-overwhelming communication to the user.

This skill implements Autonomous Optimization (AO) using Hypothesis Tree Refinement (HTR). It iteratively improves concrete artifacts (like code, models, or data pipelines) by running structured, long-term experiments against a measurable objective. It maintains a persistent hypothesis tree, coordinating subagents to test hypotheses in isolation while using a held-out test evaluator to prevent overfitting and ensure reliable, cumulative research progress. Ideal for complex benchmark tuning, agent refinement, or optimizing entire workflows.

Max is a specialized agent designed for safe, non-behavioral code improvement. It systematically refactors existing, working code by optimizing time complexity (Big-O), abstracting duplicated logic, removing dead code, and enhancing readability. It operates under a strict constraint: no change may alter the original functionality, requiring all refactored code to pass existing test suites flawlessly.

Rex is a foundational agent designed to transform vague, high-level user intent into precise, unambiguous, and structured technical specifications. It acts as a requirements analyst, asking clarifying questions, identifying edge cases, and defining scope using frameworks like MoSCoW and User Stories. It ensures that all subsequent development stages (planning, architecture, implementation) work from a single, agreed-upon artifact, minimizing guesswork and scope creep.

An advanced ruleset for crafting clear, precise, and formal English technical prose. This guide integrates established literary principles (like those from Strunk & White and Orwell) with modern best practices for AI-generated content. It is ideal for writing research papers, technical documentation, proposals, and release notes, ensuring maximum clarity and professional tone.

This skill provides a deterministic, git-only method to verify claims made by AI agents (e.g., 'done', 'shipped', 'fixed'). Instead of trusting the agent's self-report, it audits the commit's diff and git history using the DOS kernel (`dos verify`, `dos commit-audit`) to confirm if the claimed change genuinely occurred and is backed by non-forgeable evidence. It acts as a critical quality gate.