DirectiveOps
Your agents are following instructions. Bad ones cost 20% more and work worse.
ETH Zurich research found LLM-generated instruction files reduce task success by ~3% and increase inference costs 20–159%. Your files are making agents worse.
Most AGENTS.md and CLAUDE.md files are bloated, stale, or contradicting each other across tools. DirectiveOps scores instruction quality, detects drift, and helps you fix it — across every repo.
AGENTS.md is now a Linux Foundation standard; Codex, Claude Code, Cursor, Copilot, and Gemini CLI read these files. Tens of thousands of repos already ship them—few teams have fleet-wide visibility.
Connect GitHub to start, or try the demo first.
Works with
AGENTS.md is a Linux Foundation standard. Tens of thousands of repos already ship instruction files.
How it works
Coverage, drift, rollout, audit
Drift is silent; bad directives also waste cycles. Move from scattered files to inventory, governed updates, and evidence platform and security teams can rely on.
Scan instruction files and score quality — verbosity, staleness, conflicts, cross-tool consistency, and token efficiency.
Surface drift, conflicts, stale references, security findings, and cross-tool inconsistencies across repos.
Compare repos to org templates and decide the baseline platform and security teams will enforce.
Roll out updates with tracked PRs, monitor score trends, and keep an audit trail of what changed.
Quick start
Try the scanner locally
CLI scanner commands
From inside a repo, use --path . to scan the current directory.
Need fleet-wide standards and rollout? See pricing →
Open core
From scanner to fleet control
Pricing
Start free. Upgrade when the fleet needs it.
Next step
Run the scanner, try the demo, or see pricing.
View the source
Scanner, onboarding, and API
Try the hosted dashboard
Sales, support, and enterprise
Stay in the loop