AINPM.in / Work / Product artifacts / TrustVault PWA
Product artifacts · Checked 8 Nov 2026

TrustVault PWA: the product work behind it

These five artifacts show how TrustVault PWA 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.

01 · Problem brief

What problem does it solve, and for whom?

Problem

Password managers ask users to trust a server with their vault. A progressive web app can keep every secret on the device, but only if storage alone can never unlock it.

Users

Privacy-conscious individuals and families who want a password manager that works offline on any device.

Job to be done

"Keep my passwords safe on my own device, unlock with my fingerprint, and never depend on a company’s server."

Constraints

  • Zero-knowledge: a full storage dump must not reveal the vault key
  • Offline-first; works as an installable PWA
  • OWASP Mobile Top 10 (2025) as the security baseline
  • Any AI help runs locally, and no secret is ever sent to a model

Non-goals

  • Cloud sync of vault contents
  • Telemetry or hidden features
  • Remote AI services

Source: PROJECT_STATUS.md (6 Jul 2026) ↗

02 · Evaluation spec

How is quality defined and gated?

Security and quality are gated by an automated suite plus OWASP and Lighthouse checks. Figures are self-reported by the project; there is no independent audit yet.

Test set

  • About 600 unit, 400 integration and 100 security tests (crypto, auth flow, CRUD, OWASP checks)
  • Lighthouse audits for performance and PWA installability
  • OWASP Mobile Top 10 (2025) checklist, M1 to M10

Metrics

MetricMeasuredBar
Automated tests passing4,139 of 4,142>85% coverage target
TypeScript errors00
Lighthouse PWA100>90
ESLint errors in src/ · Open debt8010
WCAG 2.1 AA · PendingPartialFormal audit
Rule: Status is reported against targets, including the misses: lint errors and the accessibility audit are listed as open, not hidden.

Source: PROJECT_STATUS.md: codebase health and tests ↗

03 · Metrics tree

What does success mean, and what must not break?

My framing of metrics the project already measures.

North star

Vaults that stay usable and unreadable to anyone but their owner

A password manager fails if it leaks, and also if its owner gets locked out.

Drivers
  • Fast, safe unlock: Passkey (WebAuthn PRF) unlock with a password fallback
  • Works anywhere: Lighthouse PWA 100; offline-first caching
  • Quality holds: 4,139 of 4,142 tests passing
Guardrails
  • Storage alone never unlocks: Wrap key never stored; non-extractable session keys
  • No data egress: Zero telemetry; local-only AI with secret-free prompts
  • Breach checks leak nothing: HIBP k-anonymity (5-character hash prefix)
04 · Decision log

What was decided, and why?

DecisionDateStatusWhat and why
Phase 4 Passkey unlock without a stored key ↗10 Jun 2026AdoptedUnlock with a key derived from the authenticator’s PRF output, released only after biometric or PIN verification and never stored. Why: Zero-knowledge requires that nothing persisted can reconstruct the vault key.
Phase 6 Strict CSP and no third-party CDNs ↗15 Jun 2026AdoptedHash-based Content Security Policy with no unsafe-inline, and self-hosted OCR instead of a CDN. Why: Every external script is a way in; zero CDN egress keeps the trust boundary on the device.
Phase 6 Breach checks by k-anonymity ↗15 Jun 2026AdoptedQuery Have I Been Pwned with only a 5-character SHA-1 prefix. Why: Users get breach alerts without the password or its full hash leaving the device.
AI On-device AI behind a zero-egress boundary ↗21 Jun 2026AdoptedGemini Nano on desktop and WebLLM on Android explain password strength and breach impact locally; prompts never include secrets. Why: AI help is useful only if it cannot become a leak.
Library Extract the crypto into a reusable library ↗4 Jul 2026AdoptedMove PRF key wrapping into webauthn-prf-zktv, published on npm with a research preprint. Why: A standalone, tested library can be reviewed and reused independently of the app.
05 · Launch notes

What shipped, what's weak, and what's next?

Status: Staging deployment; feature-complete through Phase 7 plus experimental AI

Shipped

  • Credential management, generator, TOTP, auto-lock and encrypted import/export
  • Passkey unlock and multi-vault profiles
  • Breach detection and local AI explanations

Known limits

  • 801 ESLint errors in src/ to clear
  • One flaky import-export test
  • WCAG 2.1 AA formal audit pending
  • No independent security audit yet

Next

  • Full lint sweep before the next feature
  • Formal accessibility audit
  • CSV import/export and an error-boundary UI

Source: PROJECT_STATUS.md: known limitations ↗ · All product artifacts

Open it yourself: Staging app ↗ · GitHub ↗

FAQ

What else should you know about these artifacts?

How is TrustVault PWA evaluated?

Security and quality are gated by an automated suite plus OWASP and Lighthouse checks. Figures are self-reported by the project; there is no independent audit yet. Status is reported against targets, including the misses: lint errors and the accessibility audit are listed as open, not hidden.

Which decisions shaped TrustVault PWA?

Passkey unlock without a stored key (adopted, 10 Jun 2026): Zero-knowledge requires that nothing persisted can reconstruct the vault key. Strict CSP and no third-party CDNs (adopted, 15 Jun 2026): Every external script is a way in; zero CDN egress keeps the trust boundary on the device. Breach checks by k-anonymity (adopted, 15 Jun 2026): Users get breach alerts without the password or its full hash leaving the device. On-device AI behind a zero-egress boundary (adopted, 21 Jun 2026): AI help is useful only if it cannot become a leak. Extract the crypto into a reusable library (adopted, 4 Jul 2026): A standalone, tested library can be reviewed and reused independently of the app.

What are the known limits of TrustVault PWA?

801 ESLint errors in src/ to clear. One flaky import-export test. WCAG 2.1 AA formal audit pending. No independent security audit yet.