guarding-agent-directives

What problem does it solve?

This Skill prevents low-value, duplicated, or unsafe additions to agent directive files by enforcing a structured verification workflow so that only concise, actionable, and project-relevant instructions are added to CLAUDE.md, AGENTS.md, or their referenced documents.

Core Features & Use Cases

• Five-question verification: Automatically evaluates Recurrence, Non-obviousness, Novelty, Actionability, and Project-specificity with one-line reasoning for each question.
• Placement recommendation & diffs: Recommends whether content belongs in the core directive, a referenced file, or a skill-internal reference and presents an exact diff for user approval.
• Review workflow & safety gates: Scans target files for duplicates, contradictions, and superseded items, offers revision options, and requires explicit confirmation before applying changes.
• Use Case: Use when a user requests adding guidance to CLAUDE.md or when an autonomous agent attempts to modify AGENTS.md to ensure changes are necessary, precise, and safe.

Quick Start

Ask the skill to evaluate a proposed addition to CLAUDE.md, run the five verification questions, and return a pass/fail summary with placement recommendation and an editable diff for confirmation.