Why should I limit the number of agent directives?

Directive files are loaded every session. Excessive rules waste tokens and create noise, which can lead to the model ignoring important instructions or hallucinating contradictions.

What is the 5-point verification process?

The skill evaluates if a rule is recurring, non-obvious, novel (not already covered), actionable (changes behavior), and project-specific rather than universal knowledge.

How does this skill help with token efficiency?

By filtering out redundant info and suggesting specific reference files instead of bloating main files, it ensures the model only processes the most relevant context.

Can I still add a rule if the skill recommends against it?

Yes. The skill is designed to provide expert guidance, but it always offers an 'Add anyway' option to respect user authority and specific project needs.

Agent Directive Guard

Name: Agent Directive Guard
Author: ether-moon

byether-moon

0•

생산성 및 워크플로

Prevents bloat in agent instruction files by verifying every proposed rule against strict high-value criteria to maintain model performance.

Agent Directive Guard optimizes your project's CLAUDE.md, AGENTS.md, and referenced instruction files by acting as a rigorous gatekeeper for new rules. Every proposed addition is put through a five-point verification process—checking for recurrence, novelty, and actionability—to ensure only the most impactful instructions consume your context window. This prevents 'instruction bloat' that often leads to model confusion and token waste, while providing smart placement recommendations to keep your configuration organized. While the skill enforces high standards, it maintains a 'user-in-the-loop' philosophy, allowing you to override any verification failure and retain final authority over your project directives.

주요 기능

01Smart placement logic to optimize token consumption and file organization

02Conflict and redundancy auditing for existing instruction files

030 GitHub stars

04User-override mechanism for final authority over all suggestions

055-point verification framework (Recurring, Non-obvious, Novel, Actionable, Project-specific)

06Automated detection of manual or autonomous directive modifications

사용 사례

01Auditing and streamlining existing CLAUDE.md files to improve agent focus

02Managing complex multi-agent instructions by distributing rules into specific reference files

03Preventing 'obvious' or redundant coding rules from cluttering project documentation

주요 기능

01Smart placement logic to optimize token consumption and file organization

02Conflict and redundancy auditing for existing instruction files

030 GitHub stars

04User-override mechanism for final authority over all suggestions

055-point verification framework (Recurring, Non-obvious, Novel, Actionable, Project-specific)

06Automated detection of manual or autonomous directive modifications

사용 사례

01Auditing and streamlining existing CLAUDE.md files to improve agent focus

02Managing complex multi-agent instructions by distributing rules into specific reference files

03Preventing 'obvious' or redundant coding rules from cluttering project documentation