Agentic Direction
Your agents are only as good as your direction.
HyperHat scores how your engineers direct AI agents, shows exactly where it breaks down, and turns the gaps into rules your whole team follows.
Direction Index · live
Measure Agentic Direction in 30 minutes.
Take on a real coding task with a live agent in a browser IDE. HyperHat scores how you plan, direct, verify, and recover, then lets you scrub the session to see where it went well and where it drifted.
Real task, no install
Direct a live agent through a real coding brief in the browser. No install, no repository access.
Coaching on every prompt
See what each prompt did well and what to change next time.
Replay every second
Scrub the session to see where it went well and where it drifted.
Share a verified result with one link
Send your score and breakdown to anyone, no account needed to view it.
For Teams
See how well your team directs AI agents.
A 30-minute session with tailored evaluations across specialized tracks. Leads get a team profile. Every engineer gets a private, moment-by-moment breakdown of how they direct agents and exactly how to get better.
Team profile
Evaluate team strength across six core dimensions (Output Quality, Verification, Iteration Efficiency, Prompt Quality, Task Decomposition, and Recovery & Debugging) against typical engineers on HyperHat.
Rules to merge today
Suggested team rules, exported to CLAUDE.md, AGENTS.md, or .cursorrules.
Delegation map
Understand which changes can safely be handed off to agents, and under what conditions, derived from your team's actual performance data.
Proof of improvement
Benchmark again every quarter and see exactly what moved.
Executive brief
Generate a one-page summary designed specifically for VPs and CFOs.
Privacy-first design
Individual scores remain private to them. Team leads get actionable aggregate metrics.
HyperHat Co-Director · Alpha
A second brain for every engineer directing agents.
A local daemon between you and agents like Claude Code, Cursor, and Windsurf. It sharpens your direction before the agent starts, catches drift while it works, and turns repeated corrections into better rule files. Nothing leaves your machine.
Sharper briefs
Preflight checks add the missing context, team rules, and ADRs before you hit send.
Less drift, less slop
Live alerts when an agent leaves scope, touches sensitive files, or loops on the same failure.
Better every session
Repeated corrections become a pull request against your rule files.
Works with your setup
Reads CLAUDE.md and AGENTS.md. Never commits for you.
FAQ
Frequently asked questions
Direction Index vs Co-Director, what stays private, and what ships today.
What is HyperHat?
HyperHat helps engineers and teams direct AI agents better. The Direction Index measures how well you direct agents. Co-Director works beside you while you direct them. For teams, the same benchmark becomes a shared profile, a set of rules, and a delegation map.
What's the difference between the Direction Index and Co-Director?
The Direction Index is a 30 minute benchmark that shows how well you direct agents today. Co-Director runs in your terminal or IDE and improves your direction on every real task.
Is HyperHat for individuals or teams?
Both. Engineers can benchmark themselves, replay sessions, and improve over time. Teams get shared trends, gaps, and rules. Individual sessions stay private to the engineer.
Does Co-Director ship today?
The Direction Index is live and open to all engineers. Co-Director is in alpha with a small number of engineering teams.
How is Co-Director different from code review tools?
Code review tools, CI linters, and PR bots evaluate code after the tokens are already spent. Co-Director evaluates intent and injects missing context before and during agent execution, so drift gets caught while it's still cheap to correct.
Does Co-Director replace CLAUDE.md or AGENTS.md?
No. Co-Director reads your existing rule files and ADRs, surfaces them when they're relevant, and turns repeated course corrections into proposed diffs against those same files.
Which AI coding agents does Co-Director work with?
Co-Director works with Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Windsurf, and Cline, reading each agent's existing configuration rather than replacing it. The Direction Index runs entirely in the browser and doesn't require any agent installed locally.
Does HyperHat send my code or prompts to the cloud?
No. Co-Director is designed to run locally and can operate in Strict or Local Model mode with no prompts, source code, or session history sent to HyperHat's servers, only conclusions and pointers sync to the cloud. The Direction Index runs sessions inside an isolated browser environment rather than connecting to your production repository.
Can managers see my score?
Not unless you share a report with them. Managers see team results only: the typical engineer on each dimension, never one person's score. You can share one of your own reports with your team's leads, and revoke it at any time. When a benchmark closes, everyone on it sees the same team results.
How do teams measure progress?
Run a second round about 90 days later. HyperHat compares the same people and tracks, and reports no detectable change when a move is within the noise.
What happens if I don't check the AI's work during a Direction Index session?
The session is still scored, but the score is capped and flagged. Verification is a scored dimension, not a formality. With no credible evidence of checking, the score stays in Insufficient Evidence and the report is marked Unverified. Thin checking keeps it below Proficient.
Get started
Find your direction baseline.
Run the 30 minute benchmark for yourself, or bring your whole team.
