Prompting Claude Fable 5.1
Behavioral differences and prompting patterns for Claude Fable 5.1 and Claude Mythos 5.1, covering effort, progress updates, tool-call batching, conve...
AI Orchestration Plugin
Summary
fable-advisor is a Claude Code plugin that uses Fable 5.1 as an orchestrator to manage multi-model AI coding workflows. It employs an 'architect pattern' to delegate tasks to specialized models like GPT-5.6 Luna or Sol based on complexity and required reasoning effort.
Highlights
- Implements an architect pattern where Fable 5.1 manages requirements, decomposition, and final verification.
- Routes tasks to specialized 'lanes': GPT-5.6 Luna for routine tasks and GPT-5.6 Sol for high-complexity tasks.
- Allows granular control over reasoning effort levels, ranging from 'low' to 'ultra' per task.
- Optimizes cost and token usage by focusing model output on judgment and specs rather than high-volume text.
- Enables cross-vendor review by using different model families for implementation and oversight.
auto-generated
Context
Audience
Software Developers and AI Engineers building agentic coding workflows
DomainAI Orchestration
Formatopen source software plugin
Accessopen source
Source Overview
Claude Code plugin that runs the session on Fable 5.1 as an orchestrator, routing implementation to GPT-5.6 Luna (routine tasks) or GPT-5.6 Sol (complex tasks) via the OpenAI Codex CLI with per-task reasoning effort levels from low to ultra.
The architect writes specs, delegates all code, and requires a clean-context Fable 5.1 review of accumulated diffs before marking any deliverable done. Requires Claude Code 2.1.170+ with Fable 5.1 access and an authenticated Codex CLI; without Codex it degrades to advisor-only mode.
Topics
Related
Claude CodeAgentic AIMulti-Model LLMsGPT-5.6AI Coding Agents
