Audit Your Project Instructions for Legacy Tool Contradictions
Identify and fix contradictory project rules using Fable's agentic judgment.
Use Claude Fable 5 to audit your project-specific instruction files (like CLAUDE.md) against your codebase to find legacy patterns and contradictions. ⚖️
The Scenario
You have a long-running project with complex documentation and setup rules (CLAUDE.md) that were originally written for older AI models like Sonnet or Opus 3.5. You want to ensure your instructions aren't holding back the new model's capabilities.
Before & after
Manually auditing a large codebase for rule contradictions and legacy patterns requires opening dozens of files and cross-referencing documentation, often taking 4–6 hours of deep focus.
With Claude Fable 5, you can upload the entire repository and use a "conflict audit" prompt. AI flags contradictions and suggests modern architectural improvements in 3–5 minutes.
The Prompt
I am upgrading my project workflow to Claude Fable 5. I have attached my project rules (CLAUDE.md) and several core source files. Please perform an 'agentic audit': identify any instructions in my rules that are now obsolete, find contradictions between my documentation and my actual code patterns, and suggest 5 high-impact optimizations that Fable 5 can handle better than previous models. [ATTACH_FILES_OR_PASTE_CLAUDE_MD]
Claude Fable 5's improved 'agentic judgment' allows it to notice when your project instructions (like a CLAUDE.md file) contradict the code it is currently writing. Use this to clean up 'technical debt' in your prompt instructions that were written for older, weaker models.
Source
Claude Platform - Claude API Docs"It read my CLAUDE.md and caught it teaching the exact pattern my own quality gate bans. I hadn’t asked for a review."