Skill UI

Generates comprehensive, long-form cheat sheets for advanced Machine Learning and Large Language Model (LLM) topics. These tutorials are highly structured, including mathematical derivations, from-scratch PyTorch implementations, detailed comparison tables, and 25 categorized high-frequency interview questions (L1, L2, L3). Ideal for deep academic study and technical interview preparation.

This skill provides an end-to-end solution to transform existing REST APIs, defined by OpenAPI or Swagger specifications, into functional MCP servers. It maps every API operation to a dedicated, callable LLM tool, automatically generating robust parameter schemas (Zod) and handling authentication wiring. Use this to expose internal or third-party APIs to LLMs, providing a complete workflow from spec ingestion to local testing and cloud deployment.

This skill provides a direct, fast shell invocation to capture the current git diff (staged, working-tree, or specific file list) as raw text. It is designed to replace slower LLM agent steps that merely execute 'git diff' commands, significantly speeding up CI/CD workflows and meta-steps that require repository state analysis.

A lightweight, read-only utility that analyzes past skill usage from decision logs. It aggregates skill co-occurrence frequencies, tracks meta-skill usage statistics, and surfaces router fixture data. The resulting JSON summary is highly structured and ideal for downstream LLM consumption, enabling users to understand their workflow patterns and most frequently used skills over time.

A lightweight tool designed to fetch data from a single URL via HTTP/HTTPS using various methods (GET, POST, PUT, DELETE). It serves as a fast, direct shell wrapper, ideal for replacing complex agent steps that only require simple data retrieval. Supports configurable timeouts and response body capping, making it perfect for quick data fetching in meta-skill workflows.

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.

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.

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.

This comprehensive canvas guides teams through developing robust AI-powered features. It ensures product decisions are rigorous by covering problem definition, model selection (LLM, RAG, etc.), data governance, evaluation metrics, user experience design (including fallbacks), and responsible AI compliance. Ideal for assessing AI readiness and preventing common product failures.

This skill implements the LLM Council concept, enabling multiple open-weight language models to collaborate on complex queries. The process follows a three-stage deliberation: 1) Parallel independent responses from selected models; 2) Peer-to-peer ranking of the responses; and 3) Final synthesis by a designated Chairman LLM. Powered by Fireworks AI, it ensures deep, multi-faceted analysis and consensus building for high-stakes decision support.

This skill generates comprehensive, academic-style survey papers as a single, self-contained HTML file. By reading an anchor resource and curating research data, it structures the content and utilizes advanced LLMs (like Kimi K2.6 via Fireworks) to generate all prose, figures, and bibliographies, suitable for rapid academic literature review.

This tool provides a comprehensive security assessment framework for Retrieval-Augmented Generation (RAG) systems and vector databases. It tests critical vulnerabilities such as embedding inversion, cross-tenant data leakage, knowledge-base poisoning, and retrieval manipulation, aligning with OWASP LLM08:2025 standards. It is essential for validating the security posture of multi-tenant and vector-search applications.