webauthn-prf-zktv: the product work behind it
These five artifacts show how webauthn-prf-zktv was framed, measured and decided, not just how it was coded. Each one summarises the project's own documents and links to them, so you can check every line. Read the case study.
What problem does it solve, and for whom?
Problem
Web apps that encrypt user data need a key that a stolen database cannot reproduce. Passkeys can supply one through the WebAuthn PRF extension, but the pattern is easy to get wrong.
Users
Developers building zero-knowledge web and PWA apps; researchers reviewing passkey-based encryption.
Job to be done
"Give me a small, reviewed library that wraps a vault key under a passkey or a password, so storage alone can never unlock it."
Constraints
- One runtime dependency; native WebAuthn APIs, no wrapper library
- Only non-extractable keys; nothing secret is ever persisted
- Symmetric primitives only in the wrap path (AES-256-GCM, HKDF-SHA256, scrypt)
Non-goals
- Combined-factor (hybrid) wrapping in v1
- Android WebView PRF, which the platform does not support
Source: README and SECURITY.md ↗
How is quality defined and gated?
Correctness is pinned by a threat model with explicit guarantees, known-answer interop vectors for independent implementations, and an automated test suite.
Test set
- Primary threat: an attacker with a full storage dump; secondary: script running in the page (XSS)
- Known-answer vectors so other implementations can check compatibility
- Vitest suite run by a verify script with type-check, lint and build
Metrics
| Metric | Measured | Bar |
|---|---|---|
| Platforms with WebAuthn PRF · Android WebView unsupported | 5 of 6 | — |
| Runtime dependencies | 1 | — |
| Wrap schemes · Passkey and password | 2 | — |
| scrypt cost | N=131072, r=8, p=1 | — |
Source: SECURITY.md guarantees ↗
What does success mean, and what must not break?
My framing of metrics the project already measures.
Apps can adopt passkey-based encryption without inventing their own crypto
The library only matters if others can use it safely.
- Works on the platforms people use: PRF on 5 of 6 platforms
- Easy to adopt: npm v0.2.0, one dependency
- Independently checkable: Interop vectors and a DOI preprint
- No lockout: Password wrap kept alongside every passkey wrap
- No key export: No exportKey path in the public API
- No decrypt oracle: One generic DecryptError
What was decided, and why?
| Decision | Date | Status | What and why |
|---|---|---|---|
| Schemes Always keep a password wrap ↗ | 5 Jul 2026 | Adopted | Treat the password scheme as the safety net that makes passkey unlock safe to deploy. Why: Deleting a passkey destroys its PRF path permanently, and the data with it. |
| Deps Native WebAuthn, one dependency ↗ | 5 Jul 2026 | Adopted | Call navigator.credentials directly with strict PRF validation; depend only on a scrypt implementation. Why: Fewer dependencies mean a smaller audit surface for a crypto library. |
| Keys Non-extractable keys and a generic decrypt error ↗ | 5 Jul 2026 | Adopted | No exportKey path; one DecryptError for every failure; zeroise transient buffers. Why: Script in the page can misuse an unlocked key but cannot steal it, and errors reveal nothing. |
| Hybrid Combined-factor scheme deferred ↗ | 5 Jul 2026 | Declined | The hybrid-v1 name is reserved but not implemented in v1. Why: Ship the two well-understood schemes first. |
What shipped, what's weak, and what's next?
Status: npm v0.2.0 (5 Jul 2026); preprint with DOI
Shipped
- Core, WebAuthn and IndexedDB modules
- Threat model with guarantees and residuals
- Interop vectors and a CC BY 4.0 preprint
Known limits
- Zeroisation is best-effort in JavaScript
- PRF output passes through JS memory briefly, by spec
- No PRF in Android WebView; password scheme only there
Next
- A combined-factor scheme
- Independent review of the implementation
Source: SECURITY.md residuals ↗ · All product artifacts
Open it yourself: npm ↗ · Preprint (DOI) ↗ · GitHub ↗
What else should you know about these artifacts?
How is webauthn-prf-zktv evaluated?
Correctness is pinned by a threat model with explicit guarantees, known-answer interop vectors for independent implementations, and an automated test suite. Storage compromise alone must never be sufficient: no stored input can recompute a key, and decrypt failures give no oracle.
Which decisions shaped webauthn-prf-zktv?
Always keep a password wrap (adopted, 5 Jul 2026): Deleting a passkey destroys its PRF path permanently, and the data with it. Native WebAuthn, one dependency (adopted, 5 Jul 2026): Fewer dependencies mean a smaller audit surface for a crypto library. Non-extractable keys and a generic decrypt error (adopted, 5 Jul 2026): Script in the page can misuse an unlocked key but cannot steal it, and errors reveal nothing. Combined-factor scheme deferred (declined, 5 Jul 2026): Ship the two well-understood schemes first.
What are the known limits of webauthn-prf-zktv?
Zeroisation is best-effort in JavaScript. PRF output passes through JS memory briefly, by spec. No PRF in Android WebView; password scheme only there.