Feature request: Built-in coherence check for CLAUDE.md and instruction files

Hi team! Sharing a pattern I've run into that other users might also hit as their CLAUDE.md files grow.

What I observed

Over time, my CLAUDE.md, memory files, and skill files accumulated instructions that subtly conflict with each other. For example, my deploy command appeared in two places with slightly different syntax. Which one the model follows depends on session length and what's still in context.

The interesting part: the model itself wrote most of these instructions (I let Claude update CLAUDE.md and memory files). So the conflicts were introduced silently, without me realizing it until something went wrong.

What might help

Some kind of coherence check when CLAUDE.md is updated, either automatically or as an on-demand command. Something that flags:

• Conflicting instructions across CLAUDE.md, memory, and skill files
• Instructions that are too vague to act on consistently
• Rules that would be more reliable as hooks than as prose instructions

What I did as a workaround

I built a custom /audit-instructions skill that reads all instruction sources and produces a structured report of conflicts, ambiguities, and hook candidates. Happy to share the prompt if it would be useful reference for the team.

Why I think this matters

As more people use Claude Code for complex projects, their instruction files will naturally grow. A lightweight coherence check could prevent a whole class of "why isn't Claude following my instructions?" frustration.

Thanks for considering this!

Environment: Claude Code CLI on macOS, project with CLAUDE.md + 10 memory files + 8 skill files