/preflight
Catch a bad plan before your AI writes the code
Audit a plan against the existing codebase to catch drift, bloat, redundancy, and misalignment before you start coding.
What it does
The most expensive AI failure isn't bad code. It's good code for the wrong plan. Agents rebuild things that already exist, invent new patterns beside your established ones, and gold-plate simple features. /preflight reads the plan, reads your actual codebase, and reports where the two disagree.
It flags the plan section that duplicates an existing helper, the new approach that ignores how your app already does auth, and the three steps that could be one. You fix the plan when it's still words, which is the cheapest moment to fix anything.
What it looks like
/preflight here's the plan for the notifications feature
Three problems. Step 2 builds an email-sending module, but you already have one in lib/email.ts, so reuse it. Step 4 adds a settings table, but your existing user preferences column already covers two of its three fields. And the plan uses polling where the rest of your app uses webhooks. That's drift from your own pattern. Fixed plan is about 40% less code.
Illustrative exchange. Output adapts to your actual project.
When to type /preflight
- Your agent (or you) just wrote a plan and you want it checked before code
- A spec feels bloated and you want a second pass
- You suspect the plan rebuilds something the project already has
- The last feature drifted from what you asked, and you're not doing that again
Pairs well with
Works in every major agent
/preflight installs as a plain file in your project and runs wherever your agent does.
Get /preflight and 21 more commands
Install it once. One license gets your agent the full Boost OS menu, seven always-on guards, and a memory of your project that survives between sessions. Lifetime updates.
From $99 · 7-day refund · license key emailed instantly