Adversarial review of the relay admission surface:
- FAIL-SAFE on admission death: nostr-rs-relay fails OPEN if its gRPC admission
server is unreachable, so a dead admission = an unrestricted relay. The backend
now EXITS if the admission service stops (supervisor restarts a clean pair);
documented in RELAY.md. Pin the relay image to a digest (compose note).
- Registration-oracle protection: validate() now refuses a non-loopback
RELAY_ADMISSION_BIND unless RELAY_ADMISSION_ALLOW_PUBLIC=true (+ test).
- No N+1 on crafted events: recipient p-tags are capped (20, deduped) and resolved
in ONE batched query (db.any_registered_npub) instead of one-per-tag.
- Defence-in-depth size cap: admission denies events over max_event_bytes before
any DB work (the relay enforces its own limit too).
- Docs: RELAY.md Security section (fail-open, loopback oracle, default PoW=0 spam
footgun, client sender-verification); config.toml relay port 8088 (not 8080);
Docker Hub image; .env.example RELAY_ADMISSION_ALLOW_PUBLIC.
Note: the client-side sender-impersonation blocker is fixed separately in the
monorepo (verifying gift-wrap unwrap).
Completes the operator side of the relay integration:
- deploy/relay/: reference docker-compose + config.toml for a locked-down
nostr-rs-relay wired to the backend's gRPC admission (host networking so the
loopback admission socket stays private). Clearly marked a template to pin/
verify against the operator's nostr-rs-relay version.
- docs/operations/RELAY.md: what it is (inbox relay, not the only relay), the
admission architecture, the three write policies + NIP-13 PoW, bundled vs
external, verification, and the seam's modularity. Linked from OPERATOR_SETUP.
- .env.example: the RELAY_* block (all off/placeholder).
Backend relay integration (config → seam → policy → gRPC admission →
capabilities → NIP-05 → packaging/docs) is now complete. Client DM stack next.
create-admin-wallet registered a pending admin key but never latched setup, so on
a fresh instance it left the bootstrap half-open and a later `setup` collided with
it (setup's guard only checked for an ACTIVE admin; the cli key was pending).
- New atomic Database::create_admin_key_bootstrapping: under the admin-keys
advisory lock, insert the pending generated key AND latch `locked` (unless
already validly locked), append audit, commit — one transaction. Fails CLOSED on
a tampered latch (defers to reset-setup) instead of silently healing it.
- bootstrap_admin (setup) now refuses on ANY live admin key (active OR pending),
not just active, so setup composes with create-admin-wallet. The adopt-existing-
deployment path (users but no admin keys) is unaffected: no live key -> proceeds.
- reset_setup now takes the same advisory lock (all latch writers serialize) and
documents the live-key recovery path.
- CLI prints bootstrapped-vs-added and a terminating lost-secret recovery hint.
- Docs (README + OPERATOR_SETUP section 5): both mutually-exclusive bootstrap paths.
Adversarially reviewed (atomicity / lockout / regression — all ship). L1 test
extended: fresh-latch, no-relatch-when-locked, pending semantics, composition
guard, and tamper-refuse (fails closed + rolls back the key insert).
- .env.example: add the required at-rest identity secrets (SEED_FINGERPRINT_PEPPER,
REFRESH_TOKEN_PEPPER, IP_SALT). They were missing, so `cp .env.example .env` on a
fresh setup boot-failed closed on the first required-secret check.
- README: correct the first-admin bootstrap to `smirk-admin setup --pubkey`.
create-admin-wallet registers a pending key but does not latch, so on a fresh
instance it leaves the bootstrap uninitialized.
- docs/operations/OPERATOR_SETUP.md: full deployment guide — database, secrets,
per-chain backends, run + verify, admin bootstrap, registration policy, and
smoke-testing an instance (SMOKE_API_STYLE=namespaced).
Surfaced by a clean-room operator setup against a fresh instance.
jwinterm