
Mission 5 of 10
Threat-model your own app in an hour and leave with a one-page plan.
Most people armour the front door while the windows are open. Spend one focused hour mapping what could go wrong, then rank what to fix first.
What you'll do
Draw the boxes and arrows, and mark where data crosses a trust boundary.
Ask specific questions about inputs, secrets, tools, data, outputs and cost, then cross-check for missed categories.
Score likelihood times impact, then fill in your one-page threat model with your top five.
- Free: a pen, a page and an hour
- An extra AI section if your app calls a model
- A template you can reuse for the next app
- One real app or site you own, however small. The more real, the more useful
- An hour without interruptions, and something to write on (paper is fine; a markdown file is better)
- Rough knowledge of your own stack: where it runs, what database, which third parties it calls
Your progress0/6
- 01Why model first, and the cost of skipping it6 min
- 02Sketch the data flow and mark the trust boundaries12 min
- 03Walk the six surfaces and ask the questions18 min
- 04Cross-check with STRIDE, and the AI section10 min
- 05Rank by likelihood times impact, and write the plan10 min
- 06Keep it alive: when to revisit4 min
Why model first, and the cost of skipping it
Understand what an hour of modelling buys, so you don't treat it as paperwork.
Security work has a prioritisation problem: there are infinite things you could harden and a finite afternoon. Without a model you harden whatever you read about last, which is rarely your biggest risk.
A threat model is how you spend the afternoon on the window that's actually open.
The three questions
In order: What could go wrong? (enumerate, don't guess). How likely and how bad is each? (so you can rank). What's the cheapest control that moves the top ones? (so you act).
Everything else in this track is a control; this mission is how you decide which controls you need and in what order.
The mistake it prevents
The failure mode it prevents is real and common: a solo founder spends a weekend on a perfect CSP while the app has no login rate limit and a public storage bucket. The CSP wasn't wrong; it was seventh on a list they never wrote. One hour here produces the list.
This is OWASP A04, Insecure Design: the category that exists because you can implement every other control correctly and still be insecure if the design never considered the threat. You can't test your way out of a design that trusts the wrong thing. You model it.