precision before the prompt
You should know your codebase better than your agent does. Preflight makes sure you do — before you write the prompt, before the agent touches a single file.
You should know your codebase
better than your agent does.
There's a frustration building. Agents that confidently rewrite half your codebase and break things you didn't ask them to touch. Diffs you can't fully review. Changes in files you didn't know were involved. You typed a vague prompt, the agent took the wheel, and now you're a passenger in your own codebase.
The problem isn't the agent. It's the brief. When you don't know the blast radius before you prompt, you're guessing. The agent is guessing too — and it's faster at guessing than you are at reviewing.
Preflight gives you the map before you write the prompt. Query what a change touches. See every dependency, every flow, every place that value lands. Then direct the agent with surgical precision — not "refactor the auth flow" and pray, but a precise brief for a precise outcome.
You're the architect. The agent is the contractor. That's a different relationship with AI in your codebase — and it starts before the first token is spent.