For Codex · gpt-5.5 · faf-cli
This is a blind spot.
Don't make it yours!
What is this?
A bench that scores how grounded your Codex agent really is on your project — and the tiny file that enhances AGENTS.md instantly, deterministically and on-demand.
Why should I use it?
Choose a lane.
AI thrives on it. Structured YAML — fast facts, deterministic, a single source of truth. Instantly, on-demand.
The agent's playbook. Conventions, commands, the house style. AGENTS.md rules here.
FAF defines. MD Instructs.
Can you see your blind-spot?
AGENTS.md is doing your Context.
There are two lanes, not one.
Stay in lane.
Is FAF legit?
Receipts, not adjectives.
Measure your own — one command
Run the bench on your repo. Keep your AGENTS.md. Nothing replaced.
npx faf-cli@latest benchGet your two numbers — cold, and with a .faf.
The gap is what your agent's been guessing.
"What does it do that AGENTS.md doesn't?"
The fair question, and the honest answer isn't an argument — it's a number. AGENTS.md is a keeper, but why waste tokens and leave context to guesswork? We measured it and fixed it, so you don't have to.
The .faf adds the structured layer AGENTS.md was never meant to be: deterministic facts, same every read, IANA-registered. Run the bench, keep both files, and let the gap decide.