orionmilos0-jpg/fabletieredworkflow

This project is a drop-in configuration package for Claude Code, Anthropic's coding assistant, that sets up multiple specialized subagents which automatically use different AI models depending on the kind of work being done. The idea is that expensive, careful reasoning should only be spent where it matters, while repetitive or high volume work goes to cheaper models, cutting token costs by roughly 80 percent according to the author. It defines four roles. An architect, powered by the Fable model, plans and breaks down a task and writes handoff instructions but never writes code itself, so its context stays clean. A developer, powered by Opus, does the actual heavy implementation work including the build, fix, and retry cycle. A test writer, powered by Sonnet, handles boilerplate tests, mocks, and fixtures. A reviewer, also powered by Fable, reads the finished changes and reports back a verdict without making edits itself. There is also an optional integration with OpenAI's Codex tool as a second, independent reviewer from a different model family, meant to catch issues the main models might miss. Three practices keep the setup working well: the architect must hand workers a clear, complete brief before routing work to them, everyone updates a single shared run log file so the reviewer can see what happened without piecing it together from chat history, and finished runs get archived into dated files instead of piling up and adding to every future run's reading load. Installation copies a set of agent definition files and a slash command into either a single project's .claude folder, shared with the whole team through version control, or into the user's global Claude Code configuration for use across every project. Once installed, typing a single slash command with a task description triggers the whole pipeline automatically, or each tier can be invoked one step at a time for closer control. It requires access to Fable, Opus, and Sonnet models on the user's plan, and optionally the separately installed Codex CLI tool. This is aimed at developers already using Claude Code who want to reduce API costs on larger coding tasks.