docs: Nostr signer-login research + identity-import UX plan

Research report on the most frictionless "sign in with your Nostr signer"
flow for node login (NIP-07 extension + NIP-46 QR scan with Amber, password
always kept as fallback), and a plan for adding existing Nostr identities
(nsec import / npub watch-only / browser extension) to the Nostr Identities
screen.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
archipelago 2026-07-16 21:59:54 -04:00
parent f32c4db7e2
commit 9e264611e2
2 changed files with 177 additions and 0 deletions

View File

@ -0,0 +1,82 @@
# Add an existing Nostr identity to the node — UX & implementation plan
**Status:** plan only (2026-07-16), no code. Companion research: `docs/nostr-signer-login-research.md`.
## Where it lives
The **Nostr Identities** screen (`Web5Identities.vue`, backed by `identity.list` /
`identity.create`). Today every identity is **seed-derived** (`identity_manager.rs`
derives ed25519 + nostr keys from the BIP-39 master seed at an index). "Add existing"
introduces a second class of identity: one whose key material comes from *outside* the
seed.
## Two import kinds (both needed, different guarantees)
1. **Full import (nsec)** — the node holds the secret key. The identity behaves exactly
like a seed-derived one (can sign in embedded apps, publish, encrypt). NOT covered by
seed backup — flag it visibly and include it in the encrypted node backup.
2. **Linked signer (npub only)** — the node stores just the public key; signing is
delegated to the user's own signer (browser extension NIP-07, or a NIP-46 remote
signer later). Zero key custody; some features (background publishing) unavailable —
the UI should badge what works.
## The UX (matching the house style)
**Entry point:** next to "Create identity" on Nostr Identities, an **"Add existing"**
glass-button. Opens a modal with three tabs (same tab pattern as the send/receive
modals):
1. **Browser extension** (default when `window.nostr` exists)
- One button: "Connect with extension". Flow: `getPublicKey()` → show the npub +
resolved profile (kind-0 fetched via the node's relays: avatar, name — instant
recognition) → "Add this identity".
- Creates a **linked signer** identity. A challenge signature
(`signEvent` on a throwaway event) proves key possession before adding — never add
an unverified npub as "yours".
2. **Secret key (nsec)**
- Paste field (masked, `nsec1…` or hex), inline validation + derived npub preview
with the same kind-0 profile card before confirming.
- Scary-clear copy: "Your key will be stored on this node, encrypted at rest. It is
NOT part of your seed backup — back it up separately." Confirm step requires the
profile card to load or an explicit "add anyway".
- Creates a **full** identity.
3. **Public key (npub)** — watch-only
- Paste an npub for a linked identity without any signer attached yet (useful to
reserve the profile, upgrade to extension/NIP-46 signing later).
**After adding:** the identity appears in the same grid with a small origin badge —
`seed` / `imported` / `linked` — and the imported profile picture/name pulled from
relays. Everything else (picker in apps, rename, avatar) behaves uniformly.
**Removal:** existing delete flow; for `imported` identities the confirm dialog warns
the key is destroyed unless exported first (offer "Export nsec" in the identity's detail
sheet, gated behind password re-entry).
## Backend work
- `identity_manager.rs`: identity records gain `origin: Seed { index } | Imported |
Linked`, optional `nostr_secret_hex` absent for Linked. Storage: reuse the existing
encrypted identity file; imported secrets included in node backup.
- New RPCs:
- `identity.import-nostr` `{ nsec | npub, name?, verify_sig? }` → validates, derives
npub, rejects duplicates (same pubkey as any existing identity), returns the new
identity.
- `identity.fetch-profile` `{ pubkey }` → kind-0 lookup via `nostr_relays.rs` for the
preview card (frontend could also do this, but the node already has relay plumbing
and avoids CORS).
- `identity.nostr-sign` (used by the iframe NIP-07 bridge): for `Linked` identities
return a typed error the bridge translates into "ask the user's extension instead" —
phase 2; phase 1 simply hides linked identities from the in-app signer picker.
## Demo mode
Mock `identity.import-nostr` + `identity.fetch-profile` in mock-backend.js (canned
profile: picture + name for any pasted npub) so the whole add-existing flow is
demoable without real relays.
## Phasing
1. **Phase 1 (small):** nsec + npub tabs, origin badges, backup inclusion, mock.
2. **Phase 2:** extension tab with possession-proof + kind-0 preview cards everywhere.
3. **Phase 3:** NIP-46 remote-signer identities + login integration (shares the QR
plumbing from the signer-login work).

View File

@ -0,0 +1,95 @@
# Sign in to the node with a Nostr signer — research & recommendation
**Status:** research only (2026-07-16), no code. Companion plan: `docs/nostr-identity-import-plan.md`.
## What's already in the tree (and what it isn't)
The IndeeHub "sign in with signer" work is the *inverse* of this feature: the node acts
as a NIP-07 **provider** for embedded iframe apps, signing with node-held keys
(`useNostrBridge.ts` postMessage bridge → `identity.nostr-sign` etc., picker UI in
`NostrIdentityPicker.vue`). It never verifies an external signer — but the UI patterns
(picker modal, QR rendering) and the backend crypto are reusable:
- **`nostr-sdk 0.44` is already a core dependency** (`nostr_handshake.rs` runs a real
relay client) — schnorr event verification and NIP-46 client support are essentially
free on the Rust side.
- Auth today is single-password + optional TOTP, and TOTP already uses a **two-step
login** (`auth.login``auth.login.totp`) — the exact slot where a parallel
`auth.login.nostr.*` path fits.
- The node can host its own relay (strfry app), and the frontend already bundles `qrcode`.
## Candidate flows, ranked by friction
### A. Browser extension (NIP-07) — lowest friction on desktop (2 clicks)
Login page shows "Sign in with extension" when `window.nostr` exists. Server issues a
random challenge → extension signs a **kind 22242** auth event carrying the challenge →
server verifies signature + challenge + `created_at` freshness + that the pubkey is
enrolled → normal session cookie. ~50 lines of frontend, ~80 lines of Rust. No relay
involved at all.
### B. QR scan with a mobile signer (NIP-46 `nostrconnect://`) — the headline UX (scan + 1 tap)
1. Backend generates an ephemeral client keypair and renders a
`nostrconnect://<pubkey>?relay=<url>&secret=<rand>&perms=sign_event:22242&name=Archipelago` QR.
2. User scans with **Amber** (Android reference signer; Aegis/Nowser also scan;
nsec.app is paste-based; Alby is *not* a NIP-46 signer).
3. Phone connects to the relay, acks the secret; backend requests one
`sign_event:22242` over the encrypted NIP-46 channel, verifies, issues the session.
**Key architectural choice:** make the **Rust backend the NIP-46 client** (rust-nostr's
`nostr-connect` crate), talking to the relay over localhost — the browser only polls our
own RPC for "signer connected". No websocket/mixed-content issues in the Vue app.
**Relay topology:** no public relay is required by the spec — and public relays often
rate-limit ephemeral NIP-46 traffic. The node's own strfry is the ideal relay (private,
LAN-fast); the QR should carry a relay URL derived from the Host the browser used
(LAN IP / Tailscale IP — not `.local`, which Android often can't resolve).
**One empirical blocker to test first: does Amber accept plain `ws://` LAN relays?**
(Self-signed `wss://` will likely fail cert validation.) If not, route `wss://` through
the existing nginx/HTTPS cert story.
### C. Remembered NIP-46 session (persisted bunker pointer) — zero-tap repeat logins
Same as B but persists the pairing so future logins auto-approve. Adds state,
revocation surface, and "bunker offline = silent hang" failure modes. **Defer** — B
re-scans in ~5 seconds anyway.
## Recommendation
Ship **A + B behind one "Sign in with Nostr" button**; skip C for now. Password (+TOTP)
stays the permanent fallback — exactly as the user proposed, the signer is enrolled in a
step *after* password creation, never instead of it. The verification core is one shared
Rust function (sig + challenge + freshness + enrolled-pubkey → session).
- **Onboarding:** after the password (and seed) steps, an optional "Connect a signer"
card: QR (nostrconnect) + "Use browser extension" + Skip. Success enrolls the npub as
a login key.
- **Settings (next to TOTP):** list enrolled npubs (added date + method), "Add npub"
(paste, becomes usable after a challenge-verify), "Connect another signer" (same
QR/extension modal), "Remove" (requires password confirm; removing the last npub never
locks the account — password always works).
- **Libraries:** hand-roll the 22242 event for NIP-07 (window.nostr is a browser global);
rust-nostr `nostr-connect` for NIP-46. Avoid the 2.4 MB `nostr-login` JS bundle —
wrong fit for a self-hosted box (defaults to public bunkers); it's UX prior art only.
## Security notes
- Only pubkeys enrolled **while authenticated** (or during onboarding) may log in —
a simple `login_npubs` list next to the TOTP data in `auth.rs`.
- Challenge: 32-byte random, single-use, 25 min TTL, `created_at` ±60 s, deleted on
first verify attempt; pin an origin/host tag. Rate-limit like password attempts.
- The `secret` in the nostrconnect URI is a bearer token — one QR per attempt, expires
with the challenge.
- Policy call: signer approval should count as the second factor for TOTP accounts
(possession of phone/extension key), so nostr login doesn't silently bypass TOTP.
## Open questions
1. Amber + `ws://` LAN relay — needs a 10-minute on-device test before committing.
2. Which relay URL to embed (LAN vs Tailscale vs onion) — derive from browser Host.
3. NIP-46 encryption: spec says NIP-44, some signers still NIP-04 — rust-nostr handles
both; verify against current Amber.
4. Track draft **NIP-97 "Login with Nostr"** (matches this UX exactly, unmerged) —
align, don't depend.
**Prior art:** no mainstream self-hosted node OS (Umbrel, Start9, Alby Hub) ships Nostr
QR login for its own UI — this would be genuinely differentiating, and every building
block is already in the tree.