rayhayqal/mythos-claude-skill-forge

This repository describes an application framework built around the Claude API, designed to give developers more control over how Claude responds by applying named lens configurations that shape its behavior for different tasks. The README lists five lenses: Socratic for question-first problem decomposition, Artisan for precision and structure, Analyst for data-focused reasoning, Mentor for progressive guidance, and Mythmaker for metaphor-heavy explanations. Beyond the lens system, the README describes integration with Claude Code tooling, including pre- and post-execution hooks for validation and logging, a subagent manager that deploys parallel code-focused agents for tasks such as testing, documentation, and refactoring, and a plugin registry with hot-reload support. A marketplace for importing and exporting community-created skill packs is also described. The system is said to route between the Claude API and the OpenAI API automatically, with Claude Opus and Sonnet as primary models and OpenAI as a compatibility fallback for teams with existing infrastructure. Configuration lives in a JSON profile file in a home directory folder, specifying the active lens, fallback lens, subagent definitions with their associated hooks, marketplace extensions, and a response cache with a configurable time-to-live. The command-line interface accepts a profile name, a named lens, and a session identifier. Subagents can be invoked directly, hooks can be chained and executed, and lenses can be switched within a running session. A web dashboard, CLI, and a compact mobile UI are described as the three interface options. The README claims support for Windows, macOS, Linux, and FreeBSD natively, with iOS and Android companion apps in beta. No source code is present in the repository, it links to an external site for downloads.