What is fable-foreman?

Fable Foreman is a skill for AI coding assistants like Claude Code that turns your strongest, most expensive AI model into a team lead instead of having it do all the work itself. Once installed, it plans out a task, then hands the actual implementation work to cheaper AI workers, either Claude subagents or an OpenAI Codex CLI worker if one is installed and you agree to use it, while the stronger lead model reviews the plan and checks the results. The project's reasoning is that a top tier model doing routine implementation work is wasteful, since a cheaper model can usually do that job just as well for far less cost. It routes tasks by capability class, meaning it thinks in terms of frontier, workhorse, and fast tiers of ability rather than specific model names, so the skill keeps working correctly even as new models are released and does not need to be updated for each one. Before trusting any change as finished, the skill requires a verification step done by a separate AI instance that gets only the original task description and none of the worker's own explanation of what it did, so that a worker cannot simply convince a lazy reviewer its own flawed work is correct. Skipping this blind check is only allowed for the smallest, single file edits. When a worker fails at a task twice, the skill escalates it to a more capable tier rather than repeating the same failing attempt a third time. Installation on Claude Code involves copying a skill folder plus three companion agent definition files into Claude's configuration directories, and the project notes that both are required for full delegation and verification to work. It can also run in a more limited, single-model self review mode on Claude Desktop or claude.ai, where true delegation to separate agents is not available.