Instruction Guardian

A decision framework for routing content into the right destination and preventing re-bloat of instruction files.

Violating the letter of this skill is violating the spirit of it. If you find yourself reasoning that "this specific case is different" or "the skill didn't literally anticipate my situation, so the rule doesn't apply," stop — that reasoning is the failure mode this skill exists to catch.

When This Activates

You are about to call Edit or Write on:

• CLAUDE.md or AGENTS.md — at any path depth: the root file, or any nested instance (**/CLAUDE.md, **/AGENTS.md). Common layouts include apps/*/CLAUDE.md, packages/*/AGENTS.md, services/*/CLAUDE.md, but the trigger applies to any subdirectory location regardless of repo structure.
• MEMORY.md (append, update, or insert)
• Memory topic files under .claude/**/memory/ (e.g. ~/.claude/projects/<project-id>/memory/feedback_*.md, project_*.md, reference_*.md, user_*.md). These store the detailed content that MEMORY.md indexes — they are part of the auto-memory surface area and route by the same rules.
• Any file under .claude/rules/
• An @-import reference in any of the above

Or the user says:

• "add this to CLAUDE.md" / "document in AGENTS.md" / "update the agent instructions"
• "remember this" / "note this in the rules"