Security · DaiBX · 21 November 2026
Checking DaiBX against all 253 ASVS Level 2 requirements
A security checklist tells you what to do. A record tells you where you did it, and where you chose not to.
By Shan Padayhag · 4 min read
DaiBX is a planner I built: nested tasks, a board, a calendar and time tracking, with a Rust API on AWS Lambda and a Next.js web app. Sign-up is invite-only and every account needs an authenticator app. Before calling it secure, I wanted something better than a feeling, so I checked it against the OWASP Application Security Verification Standard, version 5.0, Level 2.
Level 1 and Level 2 together hold 253 requirements. The result is a single file in the API repository, and it describes the API as of version 1.7.0.
A record, not a checklist
A checklist invites ticking boxes. What I wanted was a record I could defend line by line, so every requirement gets one of four statuses:
- Met, with a pointer to the code or the test that proves it.
- Met, with a documented deviation: implemented differently from the letter of the requirement, with the reason written next to it.
- Not applicable, because DaiBX has no such feature.
- Gap, with a numbered fix.
The requirement numbers come straight from the OWASP repository, so anyone can check a row against the standard. Today the count is 167 met, 7 met with a documented deviation, 79 not applicable and 0 gaps.
The “not applicable” rows matter as much as the others. DaiBX has its own sign-in rather than OAuth, and no WebRTC, so those chapters are marked not applicable one row at a time, each with its reason. Saying so explicitly is how the record stays honest about its own size.
Three decisions it writes down
The rows I learned most from are the ones where the obvious answer was wrong for this app.
No account lockout
The textbook defence against password guessing is to lock an account after a few failures. The problem is that anyone who knows your email address can then lock you out of your own account.
DaiBX never locks an account. It slows the attacker down instead and answers with 429 Too Many Requests and a Retry-After header. For each account and IP address, five failures are free. After that, each failure blocks that pair for 60 seconds, doubling up to 15 minutes. On top of that sit hourly limits per account and per IP address, so spraying one account from many addresses doesn’t work either. A browser that already holds a valid remembered-device cookie skips the per-account limit, so an attacker can’t lock the owner out of their own devices.
Sessions longer than NIST suggests
NIST’s guidance for this assurance level asks people to sign in again within hours. DaiBX keeps a session for up to 7 days of inactivity and 30 days in total. The access token lives 15 minutes, and the refresh token rotates every time it’s used.
That’s a deviation, so the record says why. DaiBX holds one person’s own tasks, not money or other people’s data. Every session is stored on the server and can be ended. Any change to the account’s security asks for the password and a code again. Under those conditions, not typing a code every day is a reasonable trade. Writing that paragraph forced me to list the conditions, and each one is something the code has to keep true.
SHA-1, where it’s the standard
ASVS asks for approved hash functions, and SHA-1 isn’t one any more. DaiBX uses it in two places: authenticator codes are HMAC-SHA-1, because that’s what the authenticator-app standard and the apps themselves use, and the breached-password check looks passwords up by their SHA-1 prefix. Neither use relies on collision resistance, and stored passwords are hashed with Argon2id. So the row is “met, with a documented deviation”, not a quiet “met”.
The other deviations are smaller: a code is accepted one 30-second step early or late because phone clocks drift, and secrets live in encrypted Lambda settings rather than a separate secrets manager, to be revisited when anyone else gets access.
What the record is not
It is my own assessment. Nobody else has audited DaiBX, and the record doesn’t claim otherwise. Where it says a deviation was accepted, I’m the one who accepted it.
What it does give me is a map. When I change the sign-in flow, I know which rows to re-read. When someone asks how DaiBX handles brute force, the answer is a link, not a paragraph from memory.
The full record is public: docs/security/asvs-l2.md in dai-bx-api.