Summary This project orchestrates multiple AI models (Fable 5, Grok 4.5, GPT-5.6 Sol) for efficient, cost-effective task execution and code generation. It targets technically savvy users, offering significant benefits by using a high-capability model as an architect and cheaper models for implementation, embedding cross-vendor review and optimizing token expenditure.

How It Works

Fable Advisor uses an architect pattern within Claude Code. A premium model like Fable 5 acts as the session architect, handling requirements, decomposition, and verification. Implementation tasks route to cost-effective models like Grok 4.5 (default) or GPT-5.6 Sol (critical reasoning), drastically reducing token costs. This design minimizes expensive model token usage, focusing its power on judgment while leveraging cheaper lanes for bulk implementation.

Quick Start & Requirements

• Install: claude plugin install fable-advisor@fable-advisor via Claude plugin marketplace.
• Prerequisites: Claude Code ≥ 2.1.170; Fable 5 subscription; xAI Grok CLI installed/authenticated (grok login, x.ai/cli); optional OpenAI Codex CLI installed/authenticated (npm i -g @openai/codex, codex login).
• Usage: Start session with /model fable. For lite mode, copy agents/fable-advisor.md to ~/.claude/agents/ and use Sonnet.

Highlighted Details

• Cost Optimization: Routing minimizes expensive model token usage by delegating implementation (≈90% tokens) to cheaper lanes like Grok 4.5.
• Cross-Vendor Review: Implementation by models from different families (Grok, GPT) than the Claude architect provides inherent validation.
• Commitment Boundary Consults: fable-advisor agent acts as a read-only skeptic for architecture decisions, migrations, or complex problems at minimal cost.
• Safe Delegation: A five-part spec contract enables context-free agent delegation.

Maintenance & Community

No specific maintainer, community channel, or roadmap details are provided. The author references their blog, "Attention Heads," for deeper insights.

Licensing & Compatibility

• License: MIT.
• Compatibility: Permissive MIT license allows commercial use and integration within closed-source projects.

Limitations & Caveats

Plugin requires Claude Code environments (CLI, desktop, VS Code, web), not claude.ai. Claude model fallback is silent; external lanes (Grok, Codex) fail loudly and require separate CLI setup/authentication. GPT-5.6 Sol access may be limited during preview.