docs: correct prisma-engines + migrate-on-boot accuracy in runbook

- nix devShell uses nixos-25.05's prisma-engines 6.7.0 (has
  libquery_engine.node) — the 7.x problem is the system/unstable channel
  only, not the flake's devShell (corrects an earlier overstatement; #30
  closed as invalid).
- start.js migrate-on-boot is a no-op on nix but IS the migration path on
  docker (image ENTRYPOINT) — don't assume it's dead everywhere (#31).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Dockerfile

@@ -1,20 +1,32 @@
+# Patched from upstream kind-0/nsecbunkerd Dockerfile to use pnpm — the
+# upstream version uses `npm install` but package.json declares
+# `@nostr-dev-kit/ndk` as `workspace:*`, which only pnpm understands.
+# A clean clone of upstream fails to build with `EUNSUPPORTEDPROTOCOL`
+# under npm. Switching to pnpm matches the lockfile that ships in-repo.
+# Also drops `--frozen-lockfile` because the upstream pnpm-lock.yaml is
+# out of date vs. package.json (ERR_PNPM_OUTDATED_LOCKFILE) — bug to
+# file upstream once we've verified the rest of the stack works.
+
 FROM node:20.11-bullseye AS build
 
 WORKDIR /app
 
-# Copy package files and install dependencies
+RUN npm install -g pnpm@9
-COPY package*.json ./
+
-RUN npm install
+# Copy lockfile + manifest first so the install layer caches across
+# source changes.
+COPY package.json pnpm-lock.yaml ./
+RUN pnpm install --no-frozen-lockfile
 
 # Copy application files
 COPY . .
 
 # Generate prisma client and build the application
 RUN npx prisma generate
-RUN npm run build
+RUN pnpm run build
 
 # Runtime stage
-FROM node:20.11-alpine as runtime
+FROM node:20.11-alpine AS runtime
 
 WORKDIR /app
 
@@ -22,13 +34,25 @@ RUN apk update && \
     apk add --no-cache openssl && \
     rm -rf /var/cache/apk/*
 
+RUN npm install -g pnpm@9
+
 # Copy built files from the build stage
 COPY --from=build /app .
 
-# Install only runtime dependencies
+# Install all dependencies (including devDeps). The prisma CLI lives in
-RUN npm install --only=production
+# devDependencies but scripts/start.js invokes `prisma migrate deploy`
+# at boot, so it must be available at runtime. Dropping --prod adds the
+# CLI tooling to the runtime image — a modest size cost for the
+# correctness of the migration step.
+RUN pnpm install --no-frozen-lockfile
 
 EXPOSE 3000
 
-ENTRYPOINT [ "node", "./dist/index.js" ]
+# Run via scripts/start.js so `prisma migrate deploy` applies pending
+# migrations before the daemon spawns. The upstream Dockerfile invokes
+# ./dist/index.js directly, which silently bypasses the migration step
+# and leaves the SQLite db empty on first boot — every command that
+# touches Policy/KeyUser/Token/etc. then throws "table does not exist."
+# Caught during aiolabs/nsecbunkerd#7 diagnosis 2026-05-27.
+ENTRYPOINT [ "node", "./scripts/start.js" ]
 CMD ["start"]

docs/AUTOUNLOCK.md

@@ -0,0 +1,143 @@
+# Boot-time autounlock
+
+`nsecbunkerd` stores each managed key encrypted at rest in
+`nsecbunker.db`. By default, every key is **locked** after the daemon
+starts — clients must drive an `unlock_key` admin RPC against the
+bunker before signing / encrypting / decrypting works for that key.
+
+Autounlock is an opt-in feature that, when enabled, reads a
+passphrase from a configured source at boot and unlocks every
+non-soft-deleted key in the `Key` table automatically. This trades
+operational simplicity for a documented security weakening; read
+this whole document before enabling.
+
+## Configuration
+
+Two mutually-exclusive environment variables:
+
+| Var | Meaning |
+|---|---|
+| `NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE` | Literal passphrase string. Useful for dev / `docker compose .env` flows. |
+| `NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE_FILE` | Path to a file containing the passphrase (newline-trimmed at read). Idiomatic for sops / systemd-LoadCredential / k8s-secret / external secrets-manager flows where the passphrase comes from a separate credential store. |
+
+**If both are set, the daemon fails loud at boot** with an explicit
+error. Ambiguous config is never allowed to silently pick one.
+
+**If neither is set, autounlock is off** — behavior is identical to
+pre-#16: keys remain locked until an admin `unlock_key` RPC fires per
+key per restart.
+
+## What happens at boot when autounlock is on
+
+After the daemon's existing key-loading passes complete (unencrypted
+keys from in-process config, plain-key entries in `nsecbunker.json`),
+the autounlock pass runs:
+
+1. Read the passphrase from the configured source. Failure to read
+   (missing file, no permission) is fatal at boot.
+2. Enumerate the encrypted-at-rest entries in `nsecbunker.json`'s
+   `keys` map — entries carrying the `{iv, data}` shape from
+   `create_new_key`. Plain-key entries (`{key: ...}` shape from
+   `create_account`) are already loaded by the existing
+   `startKeys()` passes and are skipped here for log clarity.
+3. For each candidate, call `unlockKey(keyName, passphrase)`.
+   `unlockKey` is idempotent post-#16: if the key was already
+   unlocked by a prior pass, it's a no-op.
+4. Log per-key INFO on success, WARN on `unlockKey → false`
+   (typically: wrong passphrase, possibly the key was created under a
+   historical passphrase that differs from the current one), ERROR on
+   throw (typically: corrupted blob).
+5. Log one summary line:
+   `🔓 autounlock: enabled (source=<env>), unlocked N/M keys in <Xms>`.
+
+The loop is sequential — log clarity > parallelism, the unlock op
+itself is cheap (one ChaCha20 decrypt per key). For 100 keys it's
+milliseconds. If a fleet ever needs the thousands, parallelize then.
+
+The NIP-46 client channel doesn't accept RPCs that route to a key
+until that key's `Backend.start()` resolves — which happens inside
+`unlockKey`. So there's no race window where a freshly-restarted
+bunker would say "key locked" to a client while the loop is in
+flight on that key.
+
+## The security trade-off
+
+Enabling autounlock means **whoever can read the passphrase source
+can recover any key from the bunker disk.** Specifically:
+
+- The encrypt-at-rest property of `nsecbunker.db` is *preserved*
+  against `cat /var/lib/nsecbunker/*.db` alone — the database holds
+  ciphertext + IV per key, not plaintext.
+- The encrypt-at-rest property is *lost* if the attacker also has
+  access to the passphrase source. Anyone with read access to the
+  passphrase env var, the passphrase file, or the process memory at
+  the moment of autounlock can decrypt every key.
+
+This is the same trade today's deployments already make when they
+hold the passphrase in `lnbits`'s env to drive `unlock_key` RPCs
+post-restart. Autounlock makes the trade *explicit at the bunker
+level* and *visible per-deployment*, but it doesn't introduce a new
+trust requirement that didn't already exist for any deployment using
+external automation to drive unlocks.
+
+### Recommendations by deployment shape
+
+- **Dev / regtest / single-host:** literal `NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE`
+  in `docker compose .env` is fine. The threat model on a dev box
+  doesn't justify the file-source ceremony.
+- **Single-tenant production:** passphrase file on a separate
+  volume / mount with stricter access. Mount via
+  `systemd-LoadCredential` so the file is only readable by the
+  bunker process and is materialized from a sops-decrypted source
+  at boot. Avoid baking the passphrase into the container image or
+  process env list (which leaks into `ps aux`, container labels, etc.).
+- **Multi-tenant / high-security:** leave autounlock off. Orchestrate
+  unlock per-restart from an external process that prompts for the
+  passphrase out-of-band (hardware token, HSM-derived secret, human
+  approval). This preserves the property that bunker startup alone
+  doesn't restore crypto capability — a deliberate human action is
+  required.
+
+## What's *not* in scope
+
+These are deliberately out of scope for the autounlock feature.
+Separate issues to file if needed:
+
+- **Per-key passphrase support.** The current `Key` table doesn't
+  carry per-key passphrase metadata; every `create_new_key(name, passphrase)`
+  in our usage today uses the same passphrase
+  (`LNBITS_NSEC_BUNKER_KEYSTORE_PASSPHRASE`). The autounlock
+  passphrase covers every encrypted key by virtue of this
+  single-passphrase invariant. If a deployment ever needs per-key
+  passphrases, that's a separate feature (per-key passphrase-selector
+  column + per-key passphrase map).
+- **Passphrase rotation.** Re-encrypting every key under a new
+  passphrase belongs in a dedicated admin RPC (`rotate_keystore`),
+  not in autounlock.
+- **HSM / hardware-derived passphrase delivery.** Orthogonal to
+  where the passphrase comes from at unlock time — autounlock just
+  reads a string. An HSM integration would land between the
+  hardware and the file the bunker reads from.
+
+## Observability hooks
+
+The autounlock pass emits:
+
+- `🔓 autounlock: unlocked <keyName>` (INFO, one per success)
+- `⚠️  autounlock: unlockKey returned false for <keyName> ...` (WARN, one per soft failure)
+- `❌ autounlock: <keyName> failed: <err.message>` (ERROR, one per throw)
+- `🔓 autounlock: enabled (source=<env>), unlocked N/M keys in <Xms>` (summary, once)
+
+When the optional Prometheus exporter lands, counters
+`nsecbunkerd_keys_unlocked_total` and `nsecbunkerd_keys_locked_total`
+will be reported from the autounlock summary state. The current
+implementation doesn't export metrics — the log line is the
+canonical signal.
+
+## See also
+
+- `src/daemon/run.ts:Daemon.maybeAutounlock` — implementation
+- `src/daemon/run.ts:Daemon.unlockKey` — the idempotent per-key call
+- `src/daemon/admin/commands/unlock_key.ts` — the admin-RPC wrapper for manual unlock
+- aiolabs/nsecbunkerd#16 — issue with full design rationale + acceptance criteria
+- aiolabs/nsecbunkerd#15 — NDK 3.0.3 bump (the structural fix this builds on)

docs/COMPARISON-lnbits-nostr_bunker.md

@@ -0,0 +1,108 @@
+# nsecbunkerd vs. `lnbits/nostr_bunker`
+
+A comparison of this daemon (the aiolabs fork of `kind-0/nsecbunkerd`) against the
+upstream LNbits extension [`lnbits/nostr_bunker`](https://github.com/lnbits/nostr_bunker).
+
+> Source verified 2026-06-19 against `lnbits/nostr_bunker@main` (`services.py`,
+> `models.py`, `crud.py`). The two projects share a name and a NIP (NIP-46 remote
+> signing) but are architecturally **inverted**: this daemon *uses* LNbits as a
+> downstream wallet provider; the upstream extension *is* an LNbits extension that
+> turns a wallet account into the bunker.
+
+## The one thing that matters: where the nsec lives
+
+| | nsecbunkerd (this fork) | `lnbits/nostr_bunker` |
+|---|---|---|
+| **Signing key location** | On the **daemon** host, separate process from LNbits | On the **LNbits** host, inside the extension DB |
+| **At-rest protection** | Passphrase-encrypted (LND-style unlock) for manually-added keys | **Plaintext** in `nostr_bunker.bunkers_data.nsec` — no encryption |
+| **Integration direction** | LNbits is a *downstream dependency* (wallet factory) | LNbits is the *host* (wallet account = signer identity) |
+
+`crud.py:create_bunkers_data()` writes the nsec straight through
+`db.insert("nostr_bunker.bunkers_data", ...)` with no encryption step; `models.py`
+`BunkersData.nsec` is "the normalized private key stored directly." This is the exact
+posture the aiolabs roadmap (`aiolabs/lnbits#18`, "no nsec at rest on LNbits") exists
+to eliminate: the LNbits host runs extension code, payment plumbing, and a public API,
+so disk/root compromise there must NOT equal Nostr-identity compromise. The
+standalone-daemon model keeps signing off that host; the upstream extension puts the
+key right back on it, unencrypted.
+
+## Full side-by-side
+
+| Dimension | nsecbunkerd (this fork) | `lnbits/nostr_bunker` |
+|---|---|---|
+| **Form factor** | Standalone Node daemon (own process/container) | LNbits extension, runs inside the LNbits process |
+| **Stack** | TypeScript + NDK 3.0.3 + nostr-tools 2.20 + Prisma/SQLite | Python + Vue/Quasar UMD frontend |
+| **Relay transport** | Daemon opens its own relay connections (NDK); per-key kind:24133 subs pinned to explicit relays (#21) | Piggybacks the `nostrclient` extension's shared relay layer (`nostr_client.relay_manager.publish_message()`) |
+| **Tenancy** | Multi-key, multi-domain, multi-user from one daemon | One bunker per wallet account; multiplexes clients via multiple `bunker://` URLs |
+| **Admin / control plane** | Whitelisted admin npubs over E2E-encrypted Nostr events; separate bunker key holds no user key material; optional remote `app.nsecbunker.com` UI | LNbits admin UI; wallet owner is implicitly the operator |
+| **Account provisioning** | OAuth-like flow: remote `create_account` → NIP-05 file write → NIP-89 (`kind:31990`) announce → mints LNbits wallet via `usermanager` API + nostdress `lud16` | None — the LNbits account already exists; the wallet *is* the identity |
+
+## NIP-46 surface
+
+Both implement NIP-46 over kind:24133 and accept **both** NIP-04 and NIP-44 v2
+(upstream `services.py` tries `nip44_decrypt` first, falls back to `nip04_decrypt`).
+
+| Method | nsecbunkerd | `lnbits/nostr_bunker` |
+|---|---|---|
+| `connect` | ✓ | ✓ (returns secret/ack after permission check) |
+| `get_public_key` | ✓ | ✓ |
+| `sign_event` | ✓ (ACL-gated, wire-name vocab #14) | ✓ (`_assert_method_allowed` + auto/confirm flow) |
+| `nip04_encrypt` / `decrypt` | ✓ | ✓ |
+| `nip44_encrypt` / `decrypt` | ✓ | ✓ |
+| `ping` | ✓ | ✓ (`pong`) |
+| `switch_relays` | — | ✓ (returns relay list as JSON) |
+
+## Policy / permission model
+
+This is where the designs genuinely diverge, and where upstream has something worth
+borrowing.
+
+**nsecbunkerd** — relational ACL across several tables:
+- `KeyUser` — a (keyName, userPubkey) grant
+- `SigningCondition` — per-method/kind/content allow rules
+- `Policy` / `PolicyRule` — reusable rule sets with per-rule `maxUsageCount` + expiry
+- `Token` — redeemable connection grant bound to a policy, with `redeemedAt` / `revokedAt`
+- Live-policy auth re-evaluated at request time (#11)
+
+**`lnbits/nostr_bunker`** — policy is **the `bunker://` URL itself**. Each `UrlData`
+row carries its own:
+- `relays`, `secret`, `client_pubkey`
+- `permissions` (e.g. `sign_event:{kind}`), `can_read`, `can_write`
+- `auto_sign` (default `False`) vs `confirm_sign` (default `True`)
+- `expires_at`
+- `post_rate_limit_per_day` — daily cap on kind:1, enforced by counting
+  `get_signing_requests_since()` over 24h (`_assert_post_rate_limit`)
+
+Pending approvals live in `SigningRequest` (status: pending/approved/signed/rejected/error),
+mirroring this fork's `Request` + manual-approval flow.
+
+**Takeaway:** upstream's "one bunker, many scoped URLs, each URL is a self-contained
+grant" is arguably cleaner than this fork's `Token`+`Policy`+`SigningCondition` triad
+for the common case of "issue a narrowly-scoped grant to one client." If the ACL surface
+here is ever simplified, that URL-as-grant model is the reference design — note in
+particular the built-in `post_rate_limit_per_day`, which this fork has no direct
+equivalent for.
+
+## Where each fits the aiolabs stack
+
+- **nsecbunkerd is the signer; LNbits is a client of it.** This is the `#18` endgame:
+  LNbits routes signing through a `RemoteBunkerSigner` over NIP-46 (the
+  protocol-over-loopback boundary chosen deliberately over a Unix socket), and every
+  nsec — operator *and* server identity — is retired from the LNbits host.
+
+- **`lnbits/nostr_bunker` is the convenience inversion we're explicitly avoiding.**
+  Useful prior art for per-URL policy ergonomics, but adopting it as the *signer
+  location* would reintroduce plaintext nsec-at-rest on the payments host — the precise
+  thing `#18` is designed to kill.
+
+## Gaps to track on our side
+
+1. **OAuth-created keys are stored recoverable, not encrypted.**
+   `create_account.ts` writes `currentConfig.keys[keyName] = { key: key.privateKey }`,
+   unlike the passphrase-encrypted path the SECURITY-MODEL doc describes for
+   manually-added keys. The doc promises non-exfiltratable keys; the OAuth path doesn't
+   meet that bar. (We're still strictly better than upstream, which stores *all* nsecs
+   plaintext — but the doc/behavior gap is real.)
+
+2. **No per-grant rate limiting.** Upstream's `post_rate_limit_per_day` is a clean
+   primitive we lack. Worth considering as a `PolicyRule` field.

docs/acl-prior-art-survey.md

@@ -0,0 +1,297 @@
+# ACL prior-art survey — NIP-46 bunker implementations
+
+Source-verified survey of how other open-source NIP-46 remote signers model
+authorization and grant lifecycle, run to inform the #25 ACL redesign (enforce
+token + grant lifecycle live at sign time instead of via a materialized cache).
+
+> **Verification status.** Every claim below was read against actual source on
+> 2026-06-19 (clones at the commits noted per project). An initial automated
+> survey overstated several implementations (notably "Signet enforces all
+> lifecycle live" — false); the corrections are called out inline. Treat the
+> file:line citations as the authority, not the prose summaries.
+
+## TL;DR for the redesign
+
+- **Amber** is the one *positive* live-lifecycle template: store the absolute
+  deadline on the grant row, recompute the verdict against `now()` on every
+  request, treat the periodic sweep as cleanup only. It also time-boxes
+  *denials*, not just grants.
+- **Signet** (a fork of our own codebase) re-shipped our #24 bug — proof that
+  materializing a policy photocopy without a live join cannot enforce
+  grant-level TTL/usage. Its schema is still the best reference for the
+  token/policy decomposition (minus the one `applyToken` materialization line).
+- **FROSTR** has the cleanest *revocation decomposition* (3 independent layers)
+  and a good auditable-credential table — but enforces **no** live expiry
+  anywhere.
+- **promenade** confirms the **revoke = re-key** anti-pattern to avoid, and
+  debunks "FROST can't decrypt DMs" (it's a design choice, not a math limit).
+- **NDK** (which we embed) is a deliberately *blank* permit seam: we own 100% of
+  policy — and `get_public_key` bypasses the seam entirely (see #26).
+
+Decision unchanged: **Option D, leaning D1.** Amber = live-evaluation reference;
+Signet = schema reference; FROSTR = revocation-decomposition reference; NDK =
+confirmed blank seam.
+
+## Strategic decision: keep our fork, treat Signet as a parts donor (2026-06-19)
+
+Signet is a fork/re-architecture of the same kind-0/nsecbunkerd lineage we
+maintain, and is feature-richer on the standalone-operator surface (trust dial,
+suspension, NIP-49 at-rest, two-tier tokens, kill-switch, React dashboard,
+Android companion). We considered adopting it wholesale. **Decision: no — keep
+our fork as what we ship; lift Signet's patterns as needed.**
+
+Why:
+- **Replacing doesn't solve #25.** Signet re-ships our exact #24 (materialized
+  photocopy, no live grant-level join). We'd still have to do the live-join work
+  — after paying a migration cost.
+- **We'd lose the integration that makes it ours.** LNbits wallet provisioning
+  (`usermanager` + nostdress), the OAuth-like `create_account` flow, and being
+  the signer target for the #18 `RemoteBunkerSigner` endgame. Porting those into
+  Signet just means maintaining a fork of a more opinionated upstream.
+- **Lineage/bus-factor.** Our `master` tracks the canonical kind-0 upstream;
+  Signet is a solo-maintainer rewrite with choices we may not want (removed JWT
+  auth, Android surface). For a security-load-bearing component that's more risk,
+  not less.
+
+Why it's low-stakes either way: LNbits ↔ bunker is **NIP-46 over the wire** (the
+deliberate protocol-over-IPC choice), so the signer is substitutable by design.
+If our fork ever becomes a maintenance burden we can drop in any conformant
+NIP-46 signer (Signet, Amber-as-bunker, HSM-backed) with config-only changes —
+**not a one-way door.**
+
+Escape hatch (option 3, parked): run Signet unmodified behind the protocol. Only
+attractive if the LNbits provisioning/OAuth flows move out of the bunker into
+LNbits proper (plausible under #18), which would shrink the integration gap
+that's the main reason to stay. Revisit if #25 implementation reveals our
+daemon's NDK/relay/ACL plumbing is materially rougher than Signet's.
+
+---
+
+## A — daemon/server implementations with a real policy model
+
+### Signet — `Letdown2491/signet` (TS daemon + React UI + Kotlin companion)
+MIT, very active (v1.11.0, 2026-06). An extensive re-architecture of the same
+kind-0/nsecbunkerd codebase we maintain.
+
+- **Re-ships our #24.** `applyToken` (`nip46-backend.ts:807`) checks
+  `Token.expiresAt` once at redeem (`:895`), then materializes `policy.rules`
+  into lifecycle-free `SigningCondition` rows (`:845-862`); the sign-time path
+  (`acl.ts:checkRequestPermission`) never reads `Token` again.
+  `maxUsageCount`/`currentUsageCount` are touched only in the policy CRUD route —
+  never enforced. Same materialization-drift bug as ours.
+- **What it adds over us:** a coarse-cache layer for **subject-level** state on
+  `KeyUser` — `revokedAt`, `suspendedAt`/`suspendUntil`, `trustLevel` — read live
+  per request and invalidated on change (`invalidateAclCache`). Genuinely fixes
+  live *revoke* (our sibling spirekeeper#22). Puts revoke on **`KeyUser`, not
+  `Token`** — corroborating our revoke=subject / expiry=grant split.
+- **Trust dial** over a kind-risk classifier: `trustLevel ∈ {paranoid,
+  reasonable, full}`, `SAFE_KINDS` auto / `SENSITIVE_KINDS` (0/3/4/5/wallet/
+  auth/NIP-04) forced manual (`acl.ts:129-161`).
+- **Two-tier tokens:** one-time `ConnectionToken` (mandatory `expiresAt`,
+  validates connect but never auto-approves) vs policy-backed `Token` (atomic
+  claim `updateMany where redeemedAt:null`, `nip46-backend.ts:813`).
+- **Key-at-rest:** NIP-49 ncryptsec + AES-256-GCM envelope (PBKDF2-SHA256 @600k).
+- **Takeaway:** adopt its `KeyUser` subject-state + `Request` indexing; reject
+  its `applyToken` materialization; the `ConnectionToken`-vs-`Token` split *is*
+  D1 in schema form.
+
+### Amber — `greenart7c3/Amber` (Android, Kotlin/Room) ⭐ live-lifecycle reference
+MIT, very active (last commit 2026-06-19). Android signer (NIP-55 intents **and**
+NIP-46 over relays). Listed in tier A despite being mobile because its permission
+model is the strongest of any surveyed.
+
+- **Grant schema** (`ApplicationPermissionsEntity.kt:18-41`): unique composite
+  index over `(pkKey, type, kind, relay)` — per-(app × method × kind × relay).
+  Columns include `acceptable: Boolean`, `rememberType: Int`, `acceptUntil:
+  Long`, `rejectUntil: Long`.
+- **Expiry enforced LIVE** (the key finding): `IntentUtils.isRemembered()`
+  (`IntentUtils.kt:1087-1101`) is the per-request verdict and recomputes
+  `acceptUntil > TimeUtils.now()` / `rejectUntil > now()` fresh every call;
+  expired → returns `null` → falls through to a user prompt. Called on both the
+  NIP-46 relay path (`EventNotificationConsumer.kt:440-441`) and the NIP-55
+  intent path (`SignerProviderQuery.kt:183` etc.).
+- **The sweep is non-load-bearing.** `updateExpiredPermissions(time)`
+  (`ApplicationDao.kt:51`, exempts `rememberType <> 4`=ALWAYS) runs every 24h via
+  WorkManager — pure cleanup; correctness doesn't depend on it firing because the
+  decision is recomputed against `now()` on read.
+- **Time-boxed denials too:** `rejectUntil` means "reject for 5 min" decays back
+  to a prompt rather than a permanent no — a nicer primitive than a single
+  allow/deny flag.
+- **Wildcard-as-distinct-tier:** lookup ladder is exact-kind → all-kinds
+  (`kind IS NULL`, `getPermissionAllKinds`, `ApplicationDao.kt:87-91`); relay
+  wildcard matches `'*' OR '' OR NULL` in one query (`getWildcardRelayPermission`,
+  `:101-106`). Wildcard rows are explicitly queried, never an accidental
+  missing-WHERE match.
+- **Read-through LRU caches rows, not verdicts** (`CachingApplicationDao`) — keeps
+  the live `now()` re-check on every cache hit; invalidation is write-driven and
+  coarse per-app.
+- **Sign policies** (`ChooseSignPolicy.kt:32-45`, stored as `signPolicy: Int`):
+  `0` basic / `1` manual-per-new-app / `2` fully-auto (short-circuits to allow
+  before any row lookup, `IntentUtils.kt:1090`).
+- **Key-at-rest** (`SecureCryptoHelper.kt`): Android Keystore AES-256-GCM, 96-bit
+  IV / 128-bit tag, StrongBox-backed when available with TEE fallback and a
+  MediaTek denylist; optional app-level biometric gate.
+- **NIP-46 coverage** (`SignerType.kt`, `BunkerRequestUtils.kt:232-248`): connect,
+  sign_event, nip04/nip44 (+v3) encrypt/decrypt, get_public_key,
+  decrypt_zap_event, ping, switch_relays, sign_psbt, logout; both `bunker://` and
+  `nostrconnect://`.
+- **Steal for us:** absolute-deadline-on-row + recompute-vs-now per request;
+  time-boxed denials; wildcard as a distinct explicitly-queried tier; cache rows
+  not answers.
+
+### FROSTR — `FROSTR-ORG/igloo-server` + `bifrost` (TS, FROST k-of-n)
+MIT. igloo-server v1.2.0 (2026-05-28); bifrost v2.0.2 (2026-01-24). Threshold
+Schnorr over Nostr; igloo-server exposes the NIP-46 endpoint, bifrost is the node
+SDK.
+
+- **Three independent authorization layers** (the prize):
+  1. **App NIP-46 policy** — `Nip46Policy { methods?, kinds? }` (`db/nip46.ts:8-11`),
+     sessions keyed `(user_id, client_pubkey)` (`:92`), checked live per request
+     (`service.ts:508-509, 766-795`). No TTL/expiry. Session revoke is **explicit**
+     (`status='revoked'`, `:792-826`); per-method/kind revoke is **implicit** (flip
+     boolean false, audited at `:722-790`).
+  2. **Peer-transport policy** — per-peer directional `allowSend`/`allowReceive`
+     (`util/peer-policy.ts:3-9`, `docs/PEER_POLICIES.md`), enforced in bifrost
+     `_filter`/`get_recv_pubkeys` (`client.ts:226-245`). **Correction:** it's
+     *default-allow + explicit per-peer deny + last-layer-wins*, not "deny-override".
+  3. **Operator API auth** — keys stored SHA-256 hash+prefix with `revoked_at`
+     (checked first, timing-safe) + `last_used_at/ip` (`migrations/..._api_keys.sql`,
+     `database.ts:815-1047`); Argon2id password hashing (`config/crypto.ts:26-31`).
+- **No layer enforces live expiry.** `nip46_requests.expires_at` exists but is
+  never populated; the only time-based enforcement is the in-memory derived-key
+  vault (TTL + bounded reads + zeroize, `auth.ts:359-459`).
+- **Key-at-rest:** DB mode AES-256-GCM in SQLite, PBKDF2-HMAC-SHA256 **@600k**
+  (corrected from "~200k", `config/crypto.ts:7-11`); headless mode = plaintext env
+  (`GROUP_CRED`/`SHARE_CRED`).
+- **Distributed veto** is real at the participation level (a co-signer withholding
+  its partial below threshold blocks the sig) but the default signer auto-signs
+  (`middleware: {}`, `client.ts:55`) — realizing a veto needs a custom
+  `middleware.sign` not shipped by default.
+- **Share rotation** (recover → re-split, same group npub, old shares can't
+  combine) exists as a **bifrost SDK primitive** (`generate_dealer_package`), **not**
+  as an igloo-server endpoint; recovery reconstructs the full nsec in memory and
+  `/api/recover` even returns it over HTTP (`routes/recovery.ts:147-157`).
+- **Steal for us:** the 3-layer revocation decomposition; audit-event-on-grant-
+  change; `revoked_at`-checked-first + last-used credential table.
+
+### promenade — fiatjaf (Go, FROST coordinator + signer split)
+Off GitHub; cloned from fiatjaf's nostr-git (`relay.ngit.dev/npub180c…/promenade.git`),
+HEAD `70ff8439` 2026-06-18. NIP-46 method logic lives in the pinned dep
+`fiatjaf.com/nostr` (`nip46.DynamicSigner`).
+
+- **Architecture:** khatru coordinator-relay doubles as the NIP-46 endpoint, runs
+  the FROST ceremony, holds a transport/handler key but **no shard**
+  (`account_registration.go:44` carries only `frost.PublicKeyShard`); separate
+  signer daemons each hold one shard; m-of-n with m≤20 (`:79`). Signing-ceremony
+  kinds 26430–26434; account registration is kind **16430** (replaceable).
+- **No encrypted DMs — by choice, not by math.** `DynamicSigner` recognizes
+  `nip44_encrypt`/`nip44_decrypt`/`switch_relays` (`dynamic-signer.go`), but
+  promenade hardwires `AuthorizeEncryption → false` (`coordinator/nip46.go:167`)
+  and `GroupContext.Encrypt/Decrypt → "not implemented"` (`sign.go:288-302`).
+  README: *"destroyer of encryption."* **Correction:** threshold ECDH is NOT
+  impossible for FROST — `frost/ecdh.go` implements `CreateECDHShare` /
+  `AggregateECDHShards`; it's simply not plumbed in.
+- **ACL:** `AuthorizeSigning` per sign_event (`coordinator/nip46.go:86`); named
+  profiles `["profile", name, secret, restrictions]` where restrictions is a
+  `nostr.Filter` but only `Kinds` + `Until` are enforced (`:139-159`). The secret
+  is a reusable bearer capability.
+- **Lifecycle:** per-profile `Until` is the only time-bound; **no revoke API** —
+  dropping one capability means re-publishing the whole kind:16430 account signed
+  by the **master nsec**. The **revoke = re-key anti-pattern** to avoid.
+- **Key-at-rest:** nsec sharded client-side (never whole), but shards stored
+  **plaintext** in each signer's BoltDB (`acceptor.go:209`); coordinator/signer
+  identity keys from plaintext env.
+- **Relevance:** confirms (1) keep grant-revoke independent of key rotation, and
+  (2) for the #18 "bunker for everything" endgame, threshold-protecting the server
+  identity wouldn't *mathematically* preclude DM decryption — but keeping ECDH on a
+  separate non-threshold key is the cheaper path.
+
+---
+
+## B — library/SDK signer seams
+
+### NDK — `nostr-dev-kit/ndk` (we embed this) @ `4b86acd` (2026-04-05)
+nip46 under `core/src/signers/nip46/`.
+
+- Backend `NDKNip46Backend` (`backend/index.ts:58`), client `NDKNip46Signer`
+  (`index.ts:60`).
+- Permit seam: `Nip46PermitCallback = (params: {id, pubkey, method, params?}) =>
+  Promise<boolean>` (`backend/index.ts:29-43`), invoked via overridable
+  `pubkeyAllowed()` (`:229-231`) from each strategy.
+- **`get_public_key` bypasses the seam** — `backend/get-public-key.ts:3-11`
+  returns the pubkey with no `pubkeyAllowed` call. (rust-nostr's `approve()` wraps
+  every method including this one.) See #26.
+- Signature verified before dispatch (`index.ts:181`); strategies swappable
+  (`setStrategy`, `:156-158`).
+- `applyToken(pubkey, token)` default-throws (`:166-168`), invoked by the connect
+  handler when a token is present (`connect.ts:21-24`) — token policy is the
+  embedder's job.
+- **No** built-in scoping/kinds/rate-limit/expiry/persistence — all policy lives
+  behind the one callback. We own 100% of the policy engine.
+
+### rust-nostr / nostr-sdk @ `e47b572` (v0.45.0-alpha.1)
+- `NostrConnectRemoteSigner` (`signer.rs:39`) + `NostrConnect` client.
+- Trait `NostrConnectSignerActions::approve(&self, public_key, req) -> bool`
+  (`signer.rs:342-345`), synchronous bool, wraps the **entire** request match in
+  `serve()` (`:201-202`) — gates every method **including** `get_public_key`.
+- FFI (uniffi/wasm) exposes **only the client** `NostrConnect`, not the backend —
+  no non-Rust embedding of the signer side.
+
+### nak — `fiatjaf/nak` bunker subcommand @ `483bf94`
+- Allow-list of client pubkeys (`BunkerConfig.Clients`), `--persist`s 0600 JSON.
+- Once authorized, **signs everything** — no method/kind scoping, no expiry, no
+  rate limiting. Notably its underlying lib computes a `harmless` (connect/
+  get_public_key/ping) vs dangerous (sign/encrypt/decrypt) hint that nak
+  **discards**. A bare always-sign baseline.
+
+---
+
+## C — clients / extensions (less relevant; novel UX only)
+
+- **keys.band** — Svelte Chrome extension (NIP-07): the one browser signer with
+  *time-bounded* authorization grants (allow-for-N-minutes/session). Relevant to a
+  TTL-grant UX.
+- **nos2x / nos2x-fox** (fiatjaf) — origin of the per-origin "remember / allow
+  this site" NIP-07 model; key stored ~plaintext in extension storage.
+- **Gossip** (Rust desktop) — not a bunker, but best-in-class key-at-rest:
+  passphrase-encrypted on disk, startup unlock, memory zeroed before free. Clean
+  `LocalSigner` envelope reference.
+- **Primal**, **nowser** (Flutter) — clients that also serve NIP-46/NIP-55; use the
+  standard `optional_requested_perms` per-method/per-kind grammar.
+
+---
+
+## D — not bunkers / dead
+
+- **`Letdown2491/nip46-relay`** — a NIP-46 *transport relay* (forwards opaque
+  blobs), no signing/authz. Appears next to Signet; easy to mistake for a signer.
+- **Keychat** — Signal-over-Nostr chat app; signs only its own events.
+- **python-nostr** — abandoned 2022, no NIP-46. (No Python library offers a
+  signer-side permission abstraction; a Python bunker means hand-rolling the
+  kind-24133 loop or driving rust-nostr via FFI — and the FFI exposes only the
+  client.)
+
+---
+
+## Patterns worth stealing — consolidated
+
+1. **Live evaluation (Amber):** absolute deadline on the grant row; verdict is a
+   pure function recomputed vs `now()` per request; sweep is cleanup-only. This is
+   Option D, proven in production.
+2. **Time-box denials too (Amber `rejectUntil`):** a deny decays to a prompt.
+3. **Wildcard as a distinct, explicitly-queried tier (Amber):** never a fuzzy
+   missing-WHERE match in the auto-decide path.
+4. **Cache rows, never verdicts (Amber `CachingApplicationDao`, Signet coarse
+   cache):** keep the `now()` re-check on every hit; invalidate on write.
+5. **Subject vs grant separation (Signet):** revoke/suspend/trust on `KeyUser`
+   (cheap, cache+invalidate); expiry/usage on `Token`/`Policy` (must join live).
+6. **Usage = COUNT(Request) in window (lnbits/nostr_bunker), not a mutable
+   counter:** drop `currentUsageCount`; needs `Request.keyUserId` + index.
+7. **Revocation decomposition (FROSTR):** app-grant revoke ≠ transport quarantine ≠
+   key rotation. Never collapse grant-revoke into re-key (promenade anti-pattern).
+8. **Auditable, revocable credentials (FROSTR):** `revoked_at` checked first +
+   last-used tracking; audit-event-on-grant-change decoupled from enforcement.
+9. **Single predicate `grantIsLive(now)`** used at both redeem and sign time
+   (the discipline that prevents the original drift).
+10. **NDK seam reality:** we own all policy; design around `get_public_key`
+    bypassing `pubkeyAllowed`.

docs/runbook-migrations.md

@@ -0,0 +1,71 @@
+# nsecbunkerd migration & DB-maintenance runbook
+
+Operational notes for applying schema migrations and ACL/pairing maintenance on
+deployed, **LNbits-connected** nsecbunkerd instances.
+
+## ⚠️ Never full-wipe `nsecbunker.db` on an LNbits-connected instance
+
+The nsecbunkerd ↔ LNbits pairing is **split across both systems**:
+
+- **Bunker** (`nsecbunker.db`): per-account `KeyUser` binding (keyed by LNbits's
+  stable client pubkey) + redeemed `Token` + a shared `Policy`.
+- **LNbits** (`accounts.signer_config`, `RemoteBunkerSigner`):
+  `{token, client_nsec, policy_id}`.
+
+`RemoteBunkerSigner.sign_event()` signs **directly with the stored `client_nsec`**
+— it does NOT re-connect/re-redeem, and there is **no auto-repair** on restart or
+sign-failure. `provision()` runs only at new-account creation and mints a NEW
+npub (which changes the user's nostr identity).
+
+**Consequence of a full `nsecbunker.db` wipe:** the `KeyUser` bindings are
+deleted → every LNbits account's stored config dangles → all signing fails, and
+the only standard "repair" (`provision()`) changes identities. **Do not do it.**
+
+### Correct way to strip the #24 materialized photocopies
+
+Post-#27, token grants are evaluated live via the ACL step-4 `Token → Policy →
+PolicyRule` join, so the old materialized `SigningCondition` rows are redundant
+— and, written with `expiresAt = NULL`, they would keep granting past a token's
+expiry (silently re-opening #24 for already-paired clients). Strip them with a
+**targeted delete** that preserves the pairing:
+
+```sql
+-- Verify first.
+SELECT COUNT(*) FROM Token WHERE redeemedAt IS NOT NULL;  -- bindings to preserve
+SELECT COUNT(*) FROM SigningCondition;                    -- photocopies to strip
+
+-- Keeps KeyUser + Token + Policy intact. Live-token clients keep working
+-- untouched; only the stale photocopies are removed.
+DELETE FROM SigningCondition;
+```
+
+Run against each instance's `nsecbunker.db`. If an instance was already
+full-wiped, recover by restoring the pre-wipe `nsecbunker.db` backup, then run
+the targeted delete.
+
+> Manual-override grants (`add_signing_condition`, web-approval) also live in
+> `SigningCondition`. On an LNbits-only bunker there typically are none, so a
+> blanket `DELETE FROM SigningCondition` is safe. If an instance uses manual
+> overrides, delete only the policy-derived rows you intend to strip.
+
+## Keys are never in the DB
+
+Key material lives in `nsecbunker.json` (`keys`), never in `nsecbunker.db`. A DB
+wipe loses ACL/pairing state, never keys. LNbits holds the bunker **admin nsec**
+(`LNBITS_NSEC_BUNKER_ADMIN_NSEC`) and is the sole admin client.
+
+## Schema migrations
+
+On the **nix deploy**, migrations are applied by the deploy's `prisma migrate
+deploy`, not by the daemon — `start.js`'s `npm run prisma:migrate` step fails in
+the read-only nix store (no `npm` on PATH) and is a redundant no-op there (#31).
+On **docker**, by contrast, `start.js` IS the migration path (it's the image
+`ENTRYPOINT`), so the step is not dead everywhere. After adding a migration, make
+sure the path that applies for your target actually runs it.
+
+Prisma on NixOS: the flake pins **nixos-25.05**, whose `prisma-engines` is
+**6.7.0** (ships `libquery_engine.node`) — both the devShell and the deploy's
+`package.nix` use it, so `prisma` (and the `npm run test:integration` suite) work
+in `nix develop` out of the box. Heads-up: on **nixos-unstable / a system
+`<nixpkgs>` channel**, the bare `prisma-engines` attr is 7.x with no
+`libquery_engine.node`; don't run the repo's prisma against an unstable channel.

flake.lock

@@ -0,0 +1,27 @@
+{
+  "nodes": {
+    "nixpkgs": {
+      "locked": {
+        "lastModified": 1767313136,
+        "narHash": "sha256-16KkgfdYqjaeRGBaYsNrhPRRENs0qzkQVUooNHtoy2w=",
+        "owner": "NixOS",
+        "repo": "nixpkgs",
+        "rev": "ac62194c3917d5f474c1a844b6fd6da2db95077d",
+        "type": "github"
+      },
+      "original": {
+        "owner": "NixOS",
+        "ref": "nixos-25.05",
+        "repo": "nixpkgs",
+        "type": "github"
+      }
+    },
+    "root": {
+      "inputs": {
+        "nixpkgs": "nixpkgs"
+      }
+    }
+  },
+  "root": "root",
+  "version": 7
+}

flake.nix

@@ -0,0 +1,57 @@
+{
+  description = "nsecbunkerd — Nostr remote signing daemon (NIP-46)";
+
+  inputs.nixpkgs.url = "github:NixOS/nixpkgs/nixos-25.05";
+
+  outputs = { self, nixpkgs }:
+    let
+      systems = [ "x86_64-linux" "aarch64-linux" ];
+      forAllSystems = nixpkgs.lib.genAttrs systems;
+      pkgsFor = system: import nixpkgs { inherit system; };
+    in
+    {
+      packages = forAllSystems (system:
+        let pkgs = pkgsFor system; in
+        rec {
+          default = nsecbunkerd;
+          nsecbunkerd = pkgs.callPackage ./package.nix {
+            # nixos-unstable splits prisma-engines into versioned attrs and
+            # aliases the bare `prisma-engines` to 7.x (no libquery_engine
+            # for our 6.x client). nixos-25.05 has only the bare attr at
+            # 6.7.0. Pick whichever exists so the package builds on both.
+            prisma-engines_6 = pkgs.prisma-engines_6 or pkgs.prisma-engines;
+          };
+        }
+      );
+
+      devShells = forAllSystems (system:
+        let pkgs = pkgsFor system; in
+        {
+          default = pkgs.mkShell {
+            packages = with pkgs; [
+              nodejs_20
+              pnpm_8
+              prisma
+              prisma-engines
+              python3
+              gcc
+              pkg-config
+              openssl
+              sqlite
+            ];
+
+            shellHook = ''
+              # Point prisma at the nix-provided engines so it doesn't try to
+              # download them from binaries.prisma.sh on every install.
+              export PRISMA_QUERY_ENGINE_BINARY=${pkgs.prisma-engines}/bin/query-engine
+              export PRISMA_QUERY_ENGINE_LIBRARY=${pkgs.prisma-engines}/lib/libquery_engine.node
+              export PRISMA_SCHEMA_ENGINE_BINARY=${pkgs.prisma-engines}/bin/schema-engine
+              export PRISMA_FMT_BINARY=${pkgs.prisma-engines}/bin/prisma-fmt
+              export PRISMA_INTROSPECTION_ENGINE_BINARY=${pkgs.prisma-engines}/bin/introspection-engine
+              export PRISMA_CLIENT_ENGINE_TYPE=binary
+            '';
+          };
+        }
+      );
+    };
+}

package.json

@@ -21,6 +21,7 @@
   "scripts": {
     "build": "tsup src/index.ts; tsup src/daemon/index.ts -d dist/daemon; tsup src/client.ts -d dist/client",
     "build:client": "tsup src/client.ts -d dist/client",
+    "test": "TS_NODE_TRANSPILE_ONLY=1 node -r ts-node/register --test tests/*.test.ts",
     "prisma:generate": "npx prisma generate",
     "prisma:migrate": "npx prisma migrate deploy",
     "prisma:create": "npx prisma db push --preview-feature",
@@ -39,8 +40,8 @@
     "@fastify/view": "^8.2.0",
     "@inquirer/password": "^1.1.2",
     "@inquirer/prompts": "^1.2.3",
-    "@nostr-dev-kit/ndk": "workspace:*",
+    "@nostr-dev-kit/ndk": "3.0.3",
-    "@prisma/client": "^5.4.1",
+    "@prisma/client": "^6.19.0",
     "@scure/base": "^1.1.1",
     "@types/yargs": "^17.0.24",
     "axios": "^1.6.2",
@@ -57,7 +58,7 @@
     "isomorphic-ws": "^5.0.0",
     "lnbits": "^1.1.5",
     "lnbits-ts": "^0.0.2",
-    "nostr-tools": "^1.17.0",
+    "nostr-tools": "~2.20.0",
     "websocket-polyfill": "^0.0.3",
     "ws": "^8.13.0",
     "yargs": "^17.7.2"
@@ -65,7 +66,7 @@
   "devDependencies": {
     "@types/debug": "^4.1.8",
     "@types/node": "^18.16.18",
-    "prisma": "^5.4.1",
+    "prisma": "^6.19.0",
     "ts-node": "^10.9.1",
     "tsup": "^7.2.0",
     "typescript": "^5.1.3"

package.nix

@@ -0,0 +1,146 @@
+{
+  lib,
+  stdenv,
+  pnpm_9,
+  nodejs_20,
+  makeWrapper,
+  # Pin to prisma-engines_6 (6.19.3) — package.json's `@prisma/client` +
+  # `prisma` are at ^6.19.0. The unversioned `prisma-engines` attr is now
+  # 7.x in nixpkgs which doesn't ship libquery_engine.node, so we'd fail
+  # at postinstall.
+  prisma-engines_6,
+  openssl,
+  sqlite,
+  python311,
+  pkg-config,
+  node-gyp,
+}:
+
+let
+  prisma-engines = prisma-engines_6;
+
+  # The NDK 2.8.1 → 3.0.3 bump (commit 041f431) regenerated pnpm-lock.yaml
+  # at lockfile v9 and pinned NDK as `"3.0.3"`. Lockfile + manifest agree
+  # post-bump, so the historical patch-back-to-caret-form is no longer
+  # required. Leave the no-op shim in place as a structural anchor; if a
+  # future bump regenerates the lockfile under a non-caret manifest spec
+  # again, this is the seam where the realignment goes.
+  patchNdk = "";
+
+  prismaEnv = {
+    PRISMA_SCHEMA_ENGINE_BINARY = lib.getExe' prisma-engines "schema-engine";
+    PRISMA_QUERY_ENGINE_BINARY = lib.getExe' prisma-engines "query-engine";
+    PRISMA_QUERY_ENGINE_LIBRARY = "${prisma-engines}/lib/libquery_engine.node";
+    # Prisma 6 collapsed introspection-engine into schema-engine — the
+    # binary no longer ships in prisma-engines_6. The env var is still
+    # honored if present (drops gracefully), but pointing it at a path
+    # that doesn't exist would fail at startup.
+    PRISMA_FMT_BINARY = lib.getExe' prisma-engines "prisma-fmt";
+    PRISMA_CLIENT_ENGINE_TYPE = "binary";
+  };
+in
+stdenv.mkDerivation (finalAttrs: {
+  pname = "nsecbunkerd";
+  version = "0.10.5";
+
+  src = ./.;
+
+  pnpmDeps = pnpm_9.fetchDeps {
+    inherit (finalAttrs) pname version src;
+    fetcherVersion = 2;
+    prePnpmInstall = patchNdk;
+    hash = "sha256-DkFzzsQTuptRR8+rWfr9RGC+5XjSQrZlsZtspWfBW8w=";
+  };
+
+  postPatch = patchNdk;
+
+  nativeBuildInputs = [
+    pnpm_9.configHook
+    pnpm_9
+    nodejs_20
+    makeWrapper
+    node-gyp
+    python311
+    pkg-config
+  ];
+
+  buildInputs = [
+    openssl
+    sqlite
+  ];
+
+  env = prismaEnv;
+
+  buildPhase = ''
+    runHook preBuild
+
+    export npm_config_nodedir=${nodejs_20}
+    pnpm config set nodedir ${nodejs_20}
+
+    # configHook ran with --ignore-scripts; re-run install to trigger
+    # native-module postinstall (bcrypt). --offline keeps it inside the
+    # store seeded by configHook.
+    pnpm install --force --offline --frozen-lockfile --reporter=append-only
+
+    pnpm prisma generate
+    pnpm build
+
+    # Do NOT `pnpm prune --prod` here — the prisma CLI lives in
+    # devDependencies and `scripts/start.js` invokes it at boot via
+    # `npx prisma migrate deploy`. Without the CLI, the migration step
+    # silently fails (npx falls back to downloading prisma fresh, which
+    # OOMs on most containers) and the SQLite db stays empty. See
+    # `aiolabs/nsecbunkerd#7` diagnosis 2026-05-27.
+    find node_modules -xtype l -delete
+
+    runHook postBuild
+  '';
+
+  installPhase = ''
+    runHook preInstall
+
+    mkdir -p $out/{bin,share/nsecbunkerd}
+    # scripts/ MUST be copied — it contains the start.js launcher that
+    # runs `prisma migrate deploy` before spawning the daemon. The
+    # upstream packaging (and the upstream Dockerfile) bypassed this by
+    # invoking dist/index.js directly, leaving migrations unapplied.
+    cp -r dist node_modules prisma scripts templates package.json \
+      $out/share/nsecbunkerd/
+
+    # Wrapper invokes scripts/start.js, which runs `prisma migrate deploy`
+    # then spawns dist/index.js. start.js resolves sibling paths from
+    # __dirname, so the caller (systemd unit, docker compose, etc.) can
+    # set its own WorkingDirectory for the writable state dir without
+    # interfering with how the launcher finds its own package files.
+    # NSEC_BUNKER_CONFIG_DIR can override the config directory location;
+    # by default it's `./config` relative to cwd.
+    makeWrapper ${lib.getExe nodejs_20} $out/bin/nsecbunkerd \
+      --add-flags $out/share/nsecbunkerd/scripts/start.js \
+      --set NODE_ENV production \
+      --prefix PATH : ${lib.makeBinPath [ openssl nodejs_20 ]} \
+      ${
+        lib.concatStringsSep " \\\n      " (
+          lib.mapAttrsToList (n: v: "--set ${n} ${lib.escapeShellArg v}") prismaEnv
+        )
+      }
+
+    makeWrapper ${lib.getExe nodejs_20} $out/bin/nsecbunker-client \
+      --chdir $out/share/nsecbunkerd \
+      --add-flags $out/share/nsecbunkerd/dist/client/client.js \
+      --set NODE_ENV production
+
+    runHook postInstall
+  '';
+
+  passthru = {
+    inherit prisma-engines;
+  };
+
+  meta = {
+    description = "Nostr remote signing daemon (NIP-46)";
+    homepage = "https://github.com/kind-0/nsecbunkerd";
+    license = lib.licenses.mit;
+    mainProgram = "nsecbunkerd";
+    platforms = lib.platforms.linux;
+  };
+})

prisma/migrations/20260530112308_add_token_revoked_at/migration.sql

@@ -0,0 +1,2 @@
+-- AlterTable
+ALTER TABLE "Token" ADD COLUMN "revokedAt" DATETIME;

prisma/migrations/20260619125847_live_grant_lifecycle_schema/migration.sql

@@ -0,0 +1,37 @@
+-- RedefineTables
+PRAGMA defer_foreign_keys=ON;
+PRAGMA foreign_keys=OFF;
+CREATE TABLE "new_Request" (
+    "id" TEXT NOT NULL PRIMARY KEY,
+    "keyName" TEXT,
+    "createdAt" DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "requestId" TEXT NOT NULL,
+    "remotePubkey" TEXT NOT NULL,
+    "method" TEXT NOT NULL,
+    "params" TEXT,
+    "allowed" BOOLEAN,
+    "keyUserId" INTEGER,
+    CONSTRAINT "Request_keyUserId_fkey" FOREIGN KEY ("keyUserId") REFERENCES "KeyUser" ("id") ON DELETE SET NULL ON UPDATE CASCADE
+);
+INSERT INTO "new_Request" ("allowed", "createdAt", "id", "keyName", "method", "params", "remotePubkey", "requestId") SELECT "allowed", "createdAt", "id", "keyName", "method", "params", "remotePubkey", "requestId" FROM "Request";
+DROP TABLE "Request";
+ALTER TABLE "new_Request" RENAME TO "Request";
+CREATE INDEX "Request_keyUserId_method_idx" ON "Request"("keyUserId", "method");
+CREATE TABLE "new_SigningCondition" (
+    "id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT,
+    "method" TEXT,
+    "kind" TEXT,
+    "content" TEXT,
+    "keyUserKeyName" TEXT,
+    "allowed" BOOLEAN,
+    "keyUserId" INTEGER,
+    "createdAt" DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "expiresAt" DATETIME,
+    "revokedAt" DATETIME,
+    CONSTRAINT "SigningCondition_keyUserId_fkey" FOREIGN KEY ("keyUserId") REFERENCES "KeyUser" ("id") ON DELETE SET NULL ON UPDATE CASCADE
+);
+INSERT INTO "new_SigningCondition" ("allowed", "content", "id", "keyUserId", "keyUserKeyName", "kind", "method") SELECT "allowed", "content", "id", "keyUserId", "keyUserKeyName", "kind", "method" FROM "SigningCondition";
+DROP TABLE "SigningCondition";
+ALTER TABLE "new_SigningCondition" RENAME TO "SigningCondition";
+PRAGMA foreign_keys=ON;
+PRAGMA defer_foreign_keys=OFF;

prisma/schema.prisma

@@ -17,6 +17,14 @@ model Request {
     method        String
     params        String?
     allowed       Boolean?
+    // Bind each request to the KeyUser it was evaluated against so usage
+    // caps can be derived live by COUNTing allowed Requests, instead of
+    // maintaining a mutable PolicyRule.currentUsageCount that drifts.
+    // See aiolabs/nsecbunkerd#25 (Option D, derive-don't-count).
+    keyUserId     Int?
+    KeyUser       KeyUser? @relation(fields: [keyUserId], references: [id])
+
+    @@index([keyUserId, method])
 }
 
 model KeyUser {
@@ -31,6 +39,7 @@ model KeyUser {
   logs              Log[]
   signingConditions SigningCondition[]
   Token             Token[]
+  requests          Request[]
 
   @@unique([keyName, userPubkey], name: "unique_key_user")
 }
@@ -56,6 +65,13 @@ model User {
   pubkey    String
 }
 
+// The SigningCondition layer is the MANUAL-OVERRIDE source of truth
+// (web-approval / add_signing_condition / create_account bootstrap) — it is
+// no longer materialized from token policies (see aiolabs/nsecbunkerd#25:
+// applyToken stopped photocopying; token grants are evaluated live off
+// Token -> Policy -> PolicyRule). Under D1 the override layer carries its
+// own lifecycle so it runs through the same grantIsLive(now) predicate as
+// token grants.
 model SigningCondition {
   id             Int       @id @default(autoincrement())
   method         String?
@@ -64,6 +80,9 @@ model SigningCondition {
   keyUserKeyName String?
   allowed        Boolean?
   keyUserId      Int?
+  createdAt      DateTime  @default(now())
+  expiresAt      DateTime?
+  revokedAt      DateTime?
   KeyUser        KeyUser?  @relation(fields: [keyUserId], references: [id])
 }
 
@@ -110,6 +129,7 @@ model Token {
   deletedAt  DateTime?
   expiresAt  DateTime?
   redeemedAt DateTime?
+  revokedAt  DateTime?
   keyUserId  Int?
   policyId   Int?
   policy     Policy?   @relation(fields: [policyId], references: [id])

scripts/start.js

@@ -1,20 +1,32 @@
 const { execSync, spawn } = require('child_process');
 const fs = require('fs');
+const path = require('path');
+
+// Resolve sibling paths from this script's location so the launcher
+// works whether cwd is /app (docker), the nix store, or a writable
+// state dir set by systemd's WorkingDirectory. The prisma CLI and
+// dist/index.js live alongside this file in `<pkg>/share/nsecbunkerd/`
+// (nix) or `/app/` (docker). The migration-side env knobs:
+//   NSEC_BUNKER_CONFIG_DIR — directory holding nsecbunker.{json,db};
+//                            defaults to ./config relative to cwd.
+//   DATABASE_URL          — prisma's source of truth for the sqlite
+//                            path; honor whatever the caller set.
+const pkgRoot = path.resolve(__dirname, '..');
+const configDir = process.env.NSEC_BUNKER_CONFIG_DIR || path.resolve(process.cwd(), 'config');
 
 try {
   console.log(`Running migrations`);
-  // check if config folder exists
+  if (!fs.existsSync(configDir)) {
-  if (!fs.existsSync('./config')) {
+    fs.mkdirSync(configDir, { recursive: true });
-    execSync(`mkdir config`);
   }
-  execSync('npm run prisma:migrate');
+  execSync('npm run prisma:migrate', { cwd: pkgRoot, stdio: 'inherit' });
 } catch (error) {
   console.log(error);
   // Handle any potential migration errors here
 }
 
 const args = process.argv.slice(2);
-const childProcess = spawn('node', ['./dist/index.js', ...args], {
+const childProcess = spawn('node', [path.join(pkgRoot, 'dist/index.js'), ...args], {
   stdio: 'inherit',
 });
 

src/client.ts

@@ -100,7 +100,7 @@ function loadPrivateKey(): string | undefined {
     } else {
         // check if we have a @ so we try to get the npub from nip05
         if (remotePubkey.includes('@')) {
-            const u = await NDKUser.fromNip05(remotePubkey);
+            const u = await NDKUser.fromNip05(remotePubkey, ndk);
             if (!u) {
                 console.log(`Invalid nip05 ${remotePubkey}`);
                 process.exit(1);

src/daemon/admin/commands/add_policy_rule.ts

@@ -0,0 +1,48 @@
+import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
+import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
+import prisma from "../../../db.js";
+
+/**
+ * Add a single PolicyRule to an existing Policy. Under the live-policy
+ * auth model (#11), this propagates to every KeyUser bound to the
+ * policy at the next sign-time check.
+ *
+ * Param shape (JSON-stringified):
+ *   { policyId: number,
+ *     rule: { method: string,
+ *             kind?: number | "all" | null,
+ *             maxUsageCount?: number } }
+ *
+ * `kind` is stored as a string for parity with create_new_policy.ts's
+ * `rule.kind.toString()` storage and the override-layer convention. The
+ * `'all'` literal is honored at sign-time as a wildcard across kinds.
+ */
+export default async function addPolicyRule(admin: AdminInterface, req: NDKRpcRequest) {
+    const [ _payload ] = req.params as [ string ];
+
+    if (!_payload) throw new Error("Invalid params");
+
+    const payload = JSON.parse(_payload);
+    const { policyId, rule } = payload;
+
+    if (typeof policyId !== "number" || !rule || !rule.method) {
+        throw new Error("Invalid params");
+    }
+
+    const policy = await prisma.policy.findUnique({ where: { id: policyId } });
+    if (!policy) throw new Error("Policy not found");
+
+    await prisma.policyRule.create({
+        data: {
+            policyId,
+            method: rule.method,
+            kind: rule.kind !== undefined && rule.kind !== null ? rule.kind.toString() : null,
+            maxUsageCount: rule.maxUsageCount,
+            currentUsageCount: 0,
+        }
+    });
+
+    const result = JSON.stringify(["ok"]);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
+}

src/daemon/admin/commands/add_signing_condition.ts

@@ -0,0 +1,47 @@
+import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
+import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
+import prisma from "../../../db.js";
+
+/**
+ * Add a SigningCondition override row for a specific KeyUser. Used to
+ * grant a user a (method, kind) combination beyond the policy, or to
+ * deny one explicitly even when the policy allows.
+ *
+ * Param shape (JSON-stringified):
+ *   { keyUserId: number,
+ *     condition: { method: string,
+ *                  kind?: number | "all",
+ *                  allowed: boolean } }
+ *
+ * The override layer is consulted before the live policy join in
+ * checkIfPubkeyAllowed (step 3 vs step 4), so `allowed: false` here
+ * denies regardless of the policy.
+ */
+export default async function addSigningCondition(admin: AdminInterface, req: NDKRpcRequest) {
+    const [ _payload ] = req.params as [ string ];
+
+    if (!_payload) throw new Error("Invalid params");
+
+    const payload = JSON.parse(_payload);
+    const { keyUserId, condition } = payload;
+
+    if (typeof keyUserId !== "number" || !condition || !condition.method || typeof condition.allowed !== "boolean") {
+        throw new Error("Invalid params");
+    }
+
+    const keyUser = await prisma.keyUser.findUnique({ where: { id: keyUserId } });
+    if (!keyUser) throw new Error("KeyUser not found");
+
+    await prisma.signingCondition.create({
+        data: {
+            keyUserId,
+            method: condition.method,
+            kind: condition.kind !== undefined && condition.kind !== null ? condition.kind.toString() : null,
+            allowed: condition.allowed,
+        }
+    });
+
+    const result = JSON.stringify(["ok"]);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
+}

src/daemon/admin/commands/create_account.ts

@@ -1,4 +1,4 @@
-import { Hexpubkey, NDKKind, NDKPrivateKeySigner, NDKRpcRequest, NDKUserProfile } from "@nostr-dev-kit/ndk";
+import { Hexpubkey, NDKPrivateKeySigner, NDKRpcRequest, NDKUserProfile } from "@nostr-dev-kit/ndk";
 import AdminInterface from "..";
 import { nip19 } from 'nostr-tools';
 import { setupSkeletonProfile } from "../../lib/profile";
@@ -131,12 +131,15 @@ export default async function createAccount(admin: AdminInterface, req: NDKRpcRe
         username = payload[0];
         domain = payload[1];
         email = payload[2];
+        if (!username || !domain) {
+            throw new Error('Invalid authorization payload: missing username/domain');
+        }
         return createAccountReal(admin, req, username, domain, email);
     }
 }
 
 /**
- * This is where the real work of creating the private key, wallet, nip-05, granting access, etc happen
+ * This is where the real work of creating the private key, wallet, nip-05, granting access, etc happen — pragma: allowlist secret
  */
 export async function createAccountReal(
     admin: AdminInterface,
@@ -195,7 +198,7 @@ export async function createAccountReal(
         }
 
         const keyName = nip05;
-        const nsec = nip19.nsecEncode(key.privateKey!);
+        const nsec = key.nsec;
         currentConfig.keys[keyName] = { key: key.privateKey };
 
         saveCurrentConfig(admin.configFile, currentConfig);
@@ -209,11 +212,18 @@ export async function createAccountReal(
         // access it without having to go through an approval flow
         await grantPermissions(req, keyName);
 
-        return admin.rpc.sendResponse(req.id, req.pubkey, generatedUser.pubkey, NDKKind.NostrConnectAdmin);
+        // NDKKind.NostrConnectAdmin doesn't exist in NDK 2.8.1 — it resolves
+        // to `undefined` and sendResponse defaults to NDKKind.NostrConnect
+        // (24133), sending the response on the wrong channel. Mirror the
+        // request's kind so the response goes back on the same channel the
+        // client subscribed for. Filed as part of aiolabs/nsecbunkerd#7
+        // diagnosis 2026-05-27.
+        const originalKind = req.event.kind!;
+        return admin.rpc.sendResponse(req.id, req.pubkey, generatedUser.pubkey, originalKind);
     } catch (e: any) {
         console.trace('error', e);
-        return admin.rpc.sendResponse(req.id, req.pubkey, "error", NDKKind.NostrConnectAdmin,
+        const originalKind = req.event.kind!;
-            e.message);
+        return admin.rpc.sendResponse(req.id, req.pubkey, "error", originalKind, e.message);
     }
 }
 

src/daemon/admin/commands/create_new_key.ts

@@ -1,7 +1,7 @@
 import NDK, { NDKEvent, NDKPrivateKeySigner, NDKRpcRequest, type NostrEvent } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 import { saveEncrypted } from "../../../commands/add.js";
-import { nip19 } from 'nostr-tools';
 import { setupSkeletonProfile } from "../../lib/profile.js";
 
 export default async function createNewKey(admin: AdminInterface, req: NDKRpcRequest) {
@@ -13,7 +13,9 @@ export default async function createNewKey(admin: AdminInterface, req: NDKRpcReq
     let key;
 
     if (_nsec) {
-        key = new NDKPrivateKeySigner(nip19.decode(_nsec).data as string);
+        // NDK 3.x's `NDKPrivateKeySigner` accepts nsec1 or hex directly
+        // (see core/src/signers/private-key/index.ts `@ai-guardrail`).
+        key = new NDKPrivateKeySigner(_nsec);
     } else {
         key = NDKPrivateKeySigner.generate();
 
@@ -23,7 +25,7 @@ export default async function createNewKey(admin: AdminInterface, req: NDKRpcReq
     }
 
     const user = await key.user();
-    const nsec = nip19.nsecEncode(key.privateKey!);
+    const nsec = key.nsec;
 
     await saveEncrypted(
         admin.configFile,
@@ -38,5 +40,5 @@ export default async function createNewKey(admin: AdminInterface, req: NDKRpcReq
         npub: user.npub,
     });
 
-    return admin.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/create_new_policy.ts

@@ -1,5 +1,6 @@
 import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 import prisma from "../../../db.js";
 
 export default async function createNewPolicy(admin: AdminInterface, req: NDKRpcRequest) {
@@ -29,5 +30,5 @@ export default async function createNewPolicy(admin: AdminInterface, req: NDKRpc
     }
 
     const result = JSON.stringify(["ok"]);
-    return admin.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/create_new_token.ts

@@ -1,5 +1,6 @@
 import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 import prisma from "../../../db.js";
 
 export default async function createNewToken(admin: AdminInterface, req: NDKRpcRequest) {
@@ -7,15 +8,19 @@ export default async function createNewToken(admin: AdminInterface, req: NDKRpcR
 
     if (!clientName || !policyId) throw new Error("Invalid params");
 
-    const policy = await prisma.policy.findUnique({ where: { id: parseInt(policyId) }, include: { rules: true } });
+    const policyIdInt = parseInt(policyId);
+    const policy = await prisma.policy.findUnique({ where: { id: policyIdInt }, include: { rules: true } });
 
     if (!policy) throw new Error("Policy not found");
 
     console.log({clientName, policy, durationInHours});
 
     const token = [...Array(64)].map(() => Math.floor(Math.random() * 16).toString(16)).join('');
+    // policyId must be Int per the Prisma schema (Token.policyId references
+    // Policy.id which is autoincrement Int). Upstream passes the raw string
+    // from the wire — caught during aiolabs/nsecbunkerd#7 diagnosis 2026-05-27.
     const data: any = {
-        keyName, clientName, policyId,
+        keyName, clientName, policyId: policyIdInt,
         createdBy: req.pubkey,
         token
     };
@@ -26,5 +31,5 @@ export default async function createNewToken(admin: AdminInterface, req: NDKRpcR
     if (!tokenRecord) throw new Error("Token not created");
 
     const result = JSON.stringify(["ok"]);
-    return admin.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/ping.ts

@@ -1,6 +1,7 @@
 import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 
 export default async function ping(admin: AdminInterface, req: NDKRpcRequest) {
-    return admin.rpc.sendResponse(req.id, req.pubkey, "ok", 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, "ok", NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/remove_policy_rule.ts

@@ -0,0 +1,34 @@
+import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
+import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
+import prisma from "../../../db.js";
+
+/**
+ * Delete a single PolicyRule by id. Under the live-policy auth model
+ * (#11), removal takes effect at the next sign-time check for every
+ * KeyUser bound to that policy.
+ *
+ * Param shape (JSON-stringified):
+ *   { ruleId: number }
+ *
+ * Multi-instance bunker-sharing caveat (per comment 1473 §2):
+ * downstream reconcilers (e.g. lnbits' lnbits-default policy
+ * convergence) should treat this as an admin-only op — concurrent
+ * removes across instance versions can race. Adds are safe, removes
+ * are not.
+ */
+export default async function removePolicyRule(admin: AdminInterface, req: NDKRpcRequest) {
+    const [ _payload ] = req.params as [ string ];
+
+    if (!_payload) throw new Error("Invalid params");
+
+    const payload = JSON.parse(_payload);
+    const { ruleId } = payload;
+
+    if (typeof ruleId !== "number") throw new Error("Invalid params");
+
+    await prisma.policyRule.delete({ where: { id: ruleId } });
+
+    const result = JSON.stringify(["ok"]);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
+}

src/daemon/admin/commands/remove_signing_condition.ts

@@ -0,0 +1,27 @@
+import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
+import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
+import prisma from "../../../db.js";
+
+/**
+ * Delete a SigningCondition override row by id. Used to walk back a
+ * per-user grant or deny without affecting the policy.
+ *
+ * Param shape (JSON-stringified):
+ *   { conditionId: number }
+ */
+export default async function removeSigningCondition(admin: AdminInterface, req: NDKRpcRequest) {
+    const [ _payload ] = req.params as [ string ];
+
+    if (!_payload) throw new Error("Invalid params");
+
+    const payload = JSON.parse(_payload);
+    const { conditionId } = payload;
+
+    if (typeof conditionId !== "number") throw new Error("Invalid params");
+
+    await prisma.signingCondition.delete({ where: { id: conditionId } });
+
+    const result = JSON.stringify(["ok"]);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
+}

src/daemon/admin/commands/rename_key_user.ts

@@ -1,5 +1,6 @@
 import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 import prisma from "../../../db.js";
 
 export default async function renameKeyUser(admin: AdminInterface, req: NDKRpcRequest) {
@@ -25,5 +26,5 @@ export default async function renameKeyUser(admin: AdminInterface, req: NDKRpcRe
     });
 
     const result = JSON.stringify(["ok"]);
-    return admin.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/revoke_token.ts

@@ -0,0 +1,37 @@
+import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
+import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
+import prisma from "../../../db.js";
+
+/**
+ * Revoke a single Token without affecting other tokens for the same
+ * KeyUser. Sets Token.revokedAt to now; the live-policy auth check
+ * (#11) filters tokens with `revokedAt IS NULL` so a revoked token's
+ * policy no longer contributes to any sign-time grant for the
+ * associated KeyUser.
+ *
+ * Param shape (JSON-stringified):
+ *   { tokenId: number }
+ *
+ * The KeyUser remains intact and any other (non-revoked) tokens
+ * bound to it continue to grant via their own policies. Use
+ * revoke_user for the binary "this user is gone" case.
+ */
+export default async function revokeToken(admin: AdminInterface, req: NDKRpcRequest) {
+    const [ _payload ] = req.params as [ string ];
+
+    if (!_payload) throw new Error("Invalid params");
+
+    const payload = JSON.parse(_payload);
+    const { tokenId } = payload;
+
+    if (typeof tokenId !== "number") throw new Error("Invalid params");
+
+    await prisma.token.update({
+        where: { id: tokenId },
+        data: { revokedAt: new Date() },
+    });
+
+    const result = JSON.stringify(["ok"]);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
+}

src/daemon/admin/commands/revoke_user.ts

@@ -1,5 +1,6 @@
 import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 import prisma from "../../../db.js";
 
 export default async function revokeUser(admin: AdminInterface, req: NDKRpcRequest) {
@@ -20,5 +21,5 @@ export default async function revokeUser(admin: AdminInterface, req: NDKRpcReque
     });
 
     const result = JSON.stringify(["ok"]);
-    return admin.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/unlock_key.ts

@@ -1,5 +1,6 @@
 import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
 import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
 
 export default async function unlockKey(admin: AdminInterface, req: NDKRpcRequest) {
     const [ keyName, passphrase ] = req.params as [ string, string ];
@@ -16,5 +17,5 @@ export default async function unlockKey(admin: AdminInterface, req: NDKRpcReques
         result = JSON.stringify({ success: false, error: e.message });
     }
 
-    return admin.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
 }

src/daemon/admin/commands/update_policy.ts

@@ -0,0 +1,44 @@
+import { NDKRpcRequest } from "@nostr-dev-kit/ndk";
+import AdminInterface from "../index.js";
+import { NIP46_ADMIN_RESPONSE_KIND } from "../kinds.js";
+import prisma from "../../../db.js";
+
+/**
+ * Update mutable fields on a Policy. Currently `name` and `expiresAt`.
+ * Other Policy columns (`description`, `deletedAt`, etc.) are not
+ * exposed by this RPC — they aren't in the ratified shape, and we want
+ * a single point of intent.
+ *
+ * Param shape (JSON-stringified):
+ *   { policyId: number,
+ *     patch: { name?: string,
+ *              expiresAt?: string | null } }
+ *
+ * `expiresAt: null` explicitly clears the field; `expiresAt` absent
+ * from the patch leaves it alone.
+ */
+export default async function updatePolicy(admin: AdminInterface, req: NDKRpcRequest) {
+    const [ _payload ] = req.params as [ string ];
+
+    if (!_payload) throw new Error("Invalid params");
+
+    const payload = JSON.parse(_payload);
+    const { policyId, patch } = payload;
+
+    if (typeof policyId !== "number" || !patch || typeof patch !== "object") {
+        throw new Error("Invalid params");
+    }
+
+    const data: { name?: string; expiresAt?: Date | null } = {};
+    if (patch.name !== undefined) data.name = patch.name;
+    if (patch.expiresAt !== undefined) {
+        data.expiresAt = patch.expiresAt === null ? null : new Date(patch.expiresAt);
+    }
+
+    if (Object.keys(data).length === 0) throw new Error("Empty patch");
+
+    await prisma.policy.update({ where: { id: policyId }, data });
+
+    const result = JSON.stringify(["ok"]);
+    return admin.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
+}

src/daemon/admin/index.ts

@@ -1,5 +1,5 @@
 import "websocket-polyfill";
-import NDK, { NDKEvent, NDKKind, NDKPrivateKeySigner, NDKRpcRequest, NDKRpcResponse, NDKUser, NostrEvent } from '@nostr-dev-kit/ndk';
+import NDK, { NDKEvent, NDKKind, NDKPrivateKeySigner, NDKRpcRequest, NDKRpcResponse, NDKUser } from '@nostr-dev-kit/ndk';
 import { NDKNostrRpc } from '@nostr-dev-kit/ndk';
 import createDebug from 'debug';
 import { Key, KeyUser } from '../run';
@@ -13,11 +13,19 @@ import createNewToken from './commands/create_new_token';
 import unlockKey from './commands/unlock_key';
 import renameKeyUser from './commands/rename_key_user.js';
 import revokeUser from './commands/revoke_user';
+import addPolicyRule from './commands/add_policy_rule';
+import removePolicyRule from './commands/remove_policy_rule';
+import updatePolicy from './commands/update_policy';
+import addSigningCondition from './commands/add_signing_condition';
+import removeSigningCondition from './commands/remove_signing_condition';
+import revokeToken from './commands/revoke_token';
+import { NIP46_ADMIN_RESPONSE_KIND } from './kinds.js';
 import fs from 'fs';
 import { validateRequestFromAdmin } from './validations/request-from-admin';
 import { dmUser } from '../../utils/dm-user';
 import { IConfig, getCurrentConfig } from "../../config";
 import path from 'path';
+import { attachIndefiniteReconnect } from '../lib/relay-reconnect.js';
 
 
 const debug = createDebug("nsecbunker:admin");
@@ -55,6 +63,15 @@ class AdminInterface {
             explicitRelayUrls: opts.adminRelays,
             signer: new NDKPrivateKeySigner(opts.key),
         });
+
+        // Override NDK's "give up after detecting flapping" behavior so the
+        // bunker's admin NDK keeps trying to reconnect indefinitely. The
+        // watchdog (when enabled) still fires after 60s of zero connected
+        // relays; this helper handles shorter disconnects (e.g. an lnbits
+        // restart that pulls the nostrrelay extension's WS for a few
+        // seconds) without involving the supervisor. See aiolabs/nsecbunkerd#20.
+        attachIndefiniteReconnect(this.ndk, 'admin');
+
         this.ndk.signer?.user().then((user: NDKUser) => {
             let connectionString = `bunker://${user.npub}`;
 
@@ -111,18 +128,74 @@ class AdminInterface {
             return;
         }
 
-        this.ndk.pool.on('relay:connect', () => console.log('✅ nsecBunker Admin Interface ready'));
+        const debugTransport = process.env.NSEC_BUNKER_DEBUG_TRANSPORT === '1';
+
+        // Per-relay publish-status logging for diagnosing aiolabs/nsecbunkerd#7.
+        // NDKNostrRpc.sendResponse calls event.publish() and discards the
+        // returned Set<NDKRelay>, so a silent outbox-drop is invisible without
+        // hooking the underlying per-relay events. Gated by env flag so
+        // production deployments stay quiet.
+        const attachRelayLogging = (relay: any) => {
+            relay.on('published', (event: NDKEvent) => {
+                console.log(`📤 PUBLISHED relay=${relay.url} kind=${event.kind} id=${event.id?.slice(0,8)}`);
+            });
+            relay.on('publish:failed', (event: NDKEvent, err: any) => {
+                console.log(`❌ PUBLISH_FAILED relay=${relay.url} kind=${event.kind} id=${event.id?.slice(0,8)} err=${err?.message ?? err}`);
+            });
+        };
+
+        this.ndk.pool.on('relay:connect', (relay: any) => {
+            console.log('✅ nsecBunker Admin Interface ready');
+            if (debugTransport) attachRelayLogging(relay);
+        });
         this.ndk.pool.on('relay:disconnect', () => console.log('❌ admin disconnected'));
+
         this.ndk.connect(2500).then(() => {
             // connect for whitelisted admins
             this.rpc.subscribe({
-                "kinds": [NDKKind.NostrConnect, 24134 as number],
+                "kinds": [NDKKind.NostrConnect, NIP46_ADMIN_RESPONSE_KIND],
                 "#p": [this.signerUser!.pubkey]
             });
 
-            this.rpc.on('request', (req) => this.handleRequest(req));
+            // Attach per-relay logging to relays that connected before our
+            // 'relay:connect' listener was registered above (NDK can connect
+            // synchronously inside .connect() under some paths).
+            if (debugTransport) {
+                this.ndk.pool.relays.forEach((relay: any) => attachRelayLogging(relay));
 
-            pingOrDie(this.ndk);
+                // Wrap sendResponse to log id + kind + elapsed time so we
+                // can correlate REQUEST_IN → RESPONSE_SENT → PUBLISHED.
+                const originalSendResponse = this.rpc.sendResponse.bind(this.rpc);
+                this.rpc.sendResponse = async (id: string, remotePubkey: string, result: string, kind?: number, error?: string) => {
+                    const start = Date.now();
+                    try {
+                        await originalSendResponse(id, remotePubkey, result, kind, error);
+                        console.log(`📨 RESPONSE_SENT id=${id} remote=${remotePubkey.slice(0,8)} kind=${kind ?? NDKKind.NostrConnect} elapsed=${Date.now()-start}ms`);
+                    } catch (e: any) {
+                        console.log(`❌ RESPONSE_SEND_FAILED id=${id} remote=${remotePubkey.slice(0,8)} kind=${kind ?? NDKKind.NostrConnect} err=${e?.message ?? e}`);
+                        throw e;
+                    }
+                };
+            }
+
+            this.rpc.on('request', (req) => {
+                if (debugTransport) {
+                    console.log(`📥 REQUEST_IN method=${req.method} id=${req.id} from=${req.pubkey?.slice(0,8)} kind=${req.event?.kind}`);
+                }
+                this.handleRequest(req);
+            });
+
+            // Connection watchdog: exit if pool reports no connected relays
+            // for >60s so the process supervisor (systemd / docker restart
+            // policy / k8s) can recover. Replaces the original self-echo
+            // pingOrDie — see relayConnectionWatchdog comment + #4 + #7.
+            // Operators with external liveness checking can disable via
+            // NSEC_BUNKER_DISABLE_WATCHDOG=1.
+            if (process.env.NSEC_BUNKER_DISABLE_WATCHDOG !== '1') {
+                relayConnectionWatchdog(this.ndk);
+            } else {
+                console.log('⏸  watchdog disabled via NSEC_BUNKER_DISABLE_WATCHDOG=1');
+            }
         }).catch((err) => {
             console.log('❌ admin connection failed');
             console.log(err);
@@ -146,6 +219,12 @@ class AdminInterface {
                 case 'create_new_policy': await createNewPolicy(this, req); break;
                 case 'get_policies': await this.reqListPolicies(req); break;
                 case 'create_new_token': await createNewToken(this, req); break;
+                case 'add_policy_rule': await addPolicyRule(this, req); break;
+                case 'remove_policy_rule': await removePolicyRule(this, req); break;
+                case 'update_policy': await updatePolicy(this, req); break;
+                case 'add_signing_condition': await addSigningCondition(this, req); break;
+                case 'remove_signing_condition': await removeSigningCondition(this, req); break;
+                case 'revoke_token': await revokeToken(this, req); break;
                 default:
                     const originalKind = req.event.kind!;
                     console.log(`Unknown method ${req.method}`);
@@ -158,7 +237,15 @@ class AdminInterface {
             }
         } catch (err: any) {
             debug(`Error handling request ${req.method}: ${err?.message??err}`, req.params);
-            return this.rpc.sendResponse(req.id, req.pubkey, "error", NDKKind.NostrConnectAdmin, err?.message);
+            // NDKKind.NostrConnectAdmin doesn't exist in NDK 2.8.1 — using it
+            // makes sendResponse fall through to its default of 24133, which
+            // sends the error on a different channel than the request came in
+            // on. Mirror req.event.kind so the response goes back where the
+            // client is listening. Filed as part of aiolabs/nsecbunkerd#7
+            // diagnosis 2026-05-27.
+            const originalKind = req.event.kind!;
+            console.log(`⚠️  HANDLE_REQUEST_ERROR method=${req.method} id=${req.id} kind=${originalKind} err=${err?.message ?? err}`);
+            return this.rpc.sendResponse(req.id, req.pubkey, "error", originalKind, err?.message);
         }
     }
 
@@ -196,7 +283,7 @@ class AdminInterface {
         const key = keys.find((k) => k.name === keyName);
 
         if (!key || !key.npub) {
-            return this.rpc.sendResponse(req.id, req.pubkey, JSON.stringify([]), 24134);
+            return this.rpc.sendResponse(req.id, req.pubkey, JSON.stringify([]), NIP46_ADMIN_RESPONSE_KIND);
         }
 
         const npub = key.npub;
@@ -218,7 +305,7 @@ class AdminInterface {
             };
         }));
 
-        return this.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+        return this.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
     }
 
     /**
@@ -250,7 +337,7 @@ class AdminInterface {
             };
         }));
 
-        return this.rpc.sendResponse(req.id, req.pubkey, result, 24134);
+        return this.rpc.sendResponse(req.id, req.pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
     }
 
     /**
@@ -262,7 +349,7 @@ class AdminInterface {
         const result = JSON.stringify(await this.getKeys());
         const pubkey = req.pubkey;
 
-        return this.rpc.sendResponse(req.id, pubkey, result, 24134); // 24134
+        return this.rpc.sendResponse(req.id, pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
     }
 
     /**
@@ -274,7 +361,7 @@ class AdminInterface {
         const result = JSON.stringify(await this.getKeyUsers(req));
         const pubkey = req.pubkey;
 
-        return this.rpc.sendResponse(req.id, pubkey, result, 24134); // 24134
+        return this.rpc.sendResponse(req.id, pubkey, result, NIP46_ADMIN_RESPONSE_KIND);
     }
 
     /**
@@ -340,7 +427,7 @@ class AdminInterface {
                     remoteUser.pubkey,
                     'acl',
                     [params],
-                    24134,
+                    NIP46_ADMIN_RESPONSE_KIND,
                     (res: NDKRpcResponse) => {
                         this.requestPermissionResponse(
                             remotePubkey,
@@ -395,44 +482,47 @@ class AdminInterface {
     }
 }
 
-async function pingOrDie(ndk: NDK) {
+/**
-    let deathTimer: NodeJS.Timeout | null = null;
+ * Pool-status connection watchdog. Exits the daemon if every relay in
-    
+ * the pool stays disconnected for longer than PARTITION_THRESHOLD_MS.
-    function resetDeath() {
+ *
-        if (deathTimer) clearTimeout(deathTimer);
+ * Replaces the original `pingOrDie` self-echo watchdog, which published
-        deathTimer = setTimeout(() => {
+ * a kind-24133 event to its own pubkey every 20s and exited if it
-            console.log(`❌ No ping event received in 30 seconds. Exiting.`);
+ * didn't see the echo within 50s. That works on public relays but
-            process.exit(1);
+ * silently breaks on single-private-relay setups: NDK 2.8.1's outbox
-        }, 50000);
+ * model doesn't reliably route self-publishes back through the
-    }
+ * matching subscription, so the watchdog fires false positives and
-    
+ * exits the daemon every 50s while RPCs over the same channel still
-    const self = await ndk.signer!.user();
+ * work fine. See aiolabs/nsecbunkerd#4 + #7.
-    const sub = ndk.subscribe({
+ *
-        authors: [self.pubkey],
+ * The pool-status approach uses NDK's own connection-lifecycle
-        kinds: [NDKKind.NostrConnect],
+ * tracking — `pool.connectedRelays()` reports relays in
-        "#p": [self.pubkey]
+ * NDKRelayStatus.CONNECTED — which is reliable across all relay
-    });
+ * configurations because it doesn't depend on round-trip
-    sub.on("event", (event: NDKEvent) => {
+ * publish/subscribe. No event is published; no relay traffic.
-        console.log(`🔔 Received ping event:`, event.created_at);
+ *
-        resetDeath();
+ * Detects partition within POLL_INTERVAL + PARTITION_THRESHOLD ms.
-    });
+ * Transient disconnects shorter than PARTITION_THRESHOLD don't trip
-    sub.start();
+ * the watchdog — useful for relays that flap or briefly drop on
-
+ * network blips.
-    resetDeath();
+ */
+async function relayConnectionWatchdog(ndk: NDK) {
+    const POLL_INTERVAL_MS = 10_000;
+    const PARTITION_THRESHOLD_MS = 60_000;
+    let lastConnectedAt = Date.now();
 
     setInterval(() => {
-        const event = new NDKEvent(ndk, {
+        const connectedCount = ndk.pool.connectedRelays().length;
-            kind: NDKKind.NostrConnect,
+        if (connectedCount > 0) {
-            tags: [ ["p", self.pubkey] ],
+            lastConnectedAt = Date.now();
-            content: "ping"
+            return;
-        } as NostrEvent);
+        }
-        event.publish().then(() => {
+        const elapsed = Date.now() - lastConnectedAt;
-            console.log(`🔔 Sent ping event:`, event.created_at);
+        if (elapsed > PARTITION_THRESHOLD_MS) {
-        }).catch((e: any) => {
+            console.log(`❌ No connected relays for ${Math.floor(elapsed / 1000)}s. Exiting.`);
-            console.log(`❌ Failed to send ping event:`, e.message);
             process.exit(1);
-        });
+        }
-    }, 20000);
+    }, POLL_INTERVAL_MS);
 }
 
 export default AdminInterface;

src/daemon/admin/kinds.ts

@@ -0,0 +1,14 @@
+import type { NDKKind } from '@nostr-dev-kit/ndk';
+
+/**
+ * NIP-46 admin-RPC response channel — kind-24134. Distinct from the
+ * standard NIP-46 client channel kind-24133 (`NDKKind.NostrConnect`)
+ * which carries `sign_event` / `nip04_*` / `nip44_*` / etc.
+ *
+ * nsecbunkerd's admin surface uses a dedicated kind so signer clients
+ * and admin clients don't subscribe to each other's events.
+ *
+ * NDK 3.x's `NDKKind` enum does not include 24134; the cast happens
+ * once here so callers can pass a typed value to `rpc.sendResponse`.
+ */
+export const NIP46_ADMIN_RESPONSE_KIND = 24134 as NDKKind;

src/daemon/authorize.ts

@@ -59,9 +59,8 @@ async function createRecord(
 ) {
     let params: string | undefined;
 
-    if (param?.rawEvent) {
+    if (typeof param === 'object' && param !== null && 'rawEvent' in param) {
-        const e = param as NDKEvent;
+        params = JSON.stringify(param.rawEvent());
-        params = JSON.stringify(e.rawEvent());
     } else if (param) {
         params = param.toString();
     }

src/daemon/backend/index.ts

@@ -1,6 +1,7 @@
 import NDK, { NDKNip46Backend, NDKPrivateKeySigner, Nip46PermitCallback } from '@nostr-dev-kit/ndk';
 import prisma from '../../db.js';
 import type {FastifyInstance} from "fastify";
+import { grantIsLive } from '../lib/acl/index.js';
 
 export class Backend extends NDKNip46Backend {
     public baseUrl?: string;
@@ -18,8 +19,67 @@ export class Backend extends NDKNip46Backend {
 
         this.baseUrl = baseUrl;
         this.fastify = fastify;
+    }
 
-        // this.setStrategy('publish_event', new PublishEventHandlingStrategy());
+    /**
+     * Override NDKNip46Backend.start() to await the kind-24133
+     * subscription's EOSE before resolving. The base implementation
+     * calls `this.ndk.subscribe(...)` and returns immediately — the
+     * NDKSubscription queues a REQ on the relay connection but the
+     * relay's acknowledgement (EOSE) hasn't arrived yet. Any caller
+     * that publishes a NIP-46 event in the immediate window after
+     * `start()` returns races against the relay registering this
+     * subscription.
+     *
+     * aiolabs/lnbits#33's eager-bind chain publishes a NIP-46
+     * `connect` event in the same HTTP round-trip as `create_new_key`,
+     * which loses this race deterministically — the bunker never
+     * sees the connect event because its subscription wasn't yet
+     * registered with the relay when the event was broadcast.
+     *
+     * Awaiting EOSE closes the race: by the time `start()` resolves,
+     * the relay has confirmed it has the bunker's subscription on
+     * file and will route matching kind-24133 events to it.
+     *
+     * See aiolabs/nsecbunkerd#9 for the full diagnosis.
+     */
+    async start(): Promise<void> {
+        this.localUser = await this.signer.user();
+        await new Promise<void>((resolve) => {
+            // Pin this subscription to the daemon's explicit relays via
+            // `relayUrls`. Without that, NDK 3.x's outbox routing tries to
+            // resolve the relay set from `this.localUser.pubkey`'s NIP-65
+            // relay list (kind:10002). Newly-provisioned bunker keys have
+            // no published kind:10002 yet, so NDK's subscription manager
+            // queues the REQ waiting for a relay list that will never
+            // arrive — the kind:24133 subscription never lands on the
+            // wire, and inbound NIP-46 events (sign_event, get_public_key,
+            // nip44_*) targeted at this key get dropped by the relay
+            // with "Filter didn't match" because the bunker isn't actually
+            // subscribed for them.
+            //
+            // `relayUrls` was added in NDK 2.13.0 as the supported way to
+            // bypass outbox routing per subscription (see
+            // NDKSubscriptionOptions.relayUrls in @nostr-dev-kit/ndk).
+            // The relay set built from these URLs matches what the rest
+            // of the bunker uses (admin RPC channel + per-key Backend
+            // channels alike), so events flow through the same connection
+            // the admin interface already established.
+            //
+            // See aiolabs/nsecbunkerd#21.
+            const sub = this.ndk.subscribe(
+                {
+                    kinds: [24133],
+                    "#p": [this.localUser!.pubkey],
+                },
+                {
+                    closeOnEose: false,
+                    relayUrls: this.ndk.explicitRelayUrls,
+                }
+            );
+            sub.on("event", (e: any) => this.handleIncomingEvent(e));
+            sub.on("eose", () => resolve());
+        });
     }
 
     private async validateToken(token: string) {
@@ -32,7 +92,10 @@ export class Backend extends NDKNip46Backend {
         if (!tokenRecord) throw new Error("Token not found");
         if (tokenRecord.redeemedAt) throw new Error("Token already redeemed");
         if (!tokenRecord.policy) throw new Error("Policy not found");
-        if (tokenRecord.expiresAt && tokenRecord.expiresAt < new Date()) throw new Error("Token expired");
+        // Revoke + expiry via the single grantIsLive predicate — the exact
+        // check the sign-time ACL uses, so redeem-time and sign-time cannot
+        // drift (the root of #24). See aiolabs/nsecbunkerd#25.
+        if (!grantIsLive(tokenRecord)) throw new Error("Token expired or revoked");
 
         return tokenRecord;
     }
@@ -41,39 +104,20 @@ export class Backend extends NDKNip46Backend {
         const tokenRecord = await this.validateToken(token);
         const keyName = tokenRecord.keyName;
 
-        // Upsert the KeyUser with the given remotePubkey
+        // Record ONLY the binding (KeyUser <- Token). Under #25 the token's
+        // policy is evaluated live at sign time (checkIfPubkeyAllowed step 4)
+        // off Token -> Policy -> PolicyRule, NOT photocopied into
+        // SigningCondition rows here. That photocopy was the root of #24: the
+        // copy carried no expiry/revoke and short-circuited the live check, so
+        // an expired or revoked token kept signing forever. With no copy, the
+        // token's lifecycle is re-checked on every request and there is nothing
+        // to keep in sync.
         const upsertedUser = await prisma.keyUser.upsert({
             where: { unique_key_user: { keyName, userPubkey } },
             update: { },
             create: { keyName, userPubkey, description: tokenRecord.clientName },
         });
 
-        await prisma.signingCondition.create({
-            data: {
-                keyUserId: upsertedUser.id,
-                method: 'connect',
-                allowed: true,
-            }
-        });
-
-        // Go through the rules of this policy and apply them to the user
-        for (const rule of tokenRecord!.policy!.rules) {
-            const signingConditionQuery: any = { method: rule.method };
-
-            if (rule && rule.kind) {
-                signingConditionQuery.kind = rule.kind.toString();
-            }
-
-            await prisma.signingCondition.create({
-                data: {
-                    keyUserId: upsertedUser.id,
-                    method: rule.method,
-                    allowed: true,
-                    ...signingConditionQuery,
-                }
-            });
-        }
-
         await prisma.token.update({
             where: { id: tokenRecord.id },
             data: {

src/daemon/backend/publish-event.ts

@@ -1,14 +0,0 @@
-import { NDKNip46Backend } from "@nostr-dev-kit/ndk";
-import { IEventHandlingStrategy } from '@nostr-dev-kit/ndk';
-
-export default class PublishEventHandlingStrategy implements IEventHandlingStrategy {
-    async handle(backend: NDKNip46Backend, id: string, remotePubkey: string, params: string[]): Promise<string|undefined> {
-        const event = await backend.signEvent(remotePubkey, params);
-        if (!event) return undefined;
-
-        console.log('Publishing event', event);
-        await event.publish();
-
-        return JSON.stringify(await event.toNostrEvent());
-    }
-}

src/daemon/lib/acl/index.ts

@@ -1,13 +1,45 @@
-import { NDKEvent, NostrEvent } from '@nostr-dev-kit/ndk';
+import { NostrEvent, NIP46Method } from '@nostr-dev-kit/ndk';
 import prisma from '../../../db.js';
+import { liveWhere } from './lifecycle.js';
 
+// Re-export the single lifecycle predicate so callers (e.g.
+// Backend.validateToken) import it from the ACL module. The implementation
+// lives in ./lifecycle.ts so it can be unit-tested without a database.
+export { grantIsLive } from './lifecycle.js';
+
+/**
+ * Layered authorization check. Order matters (denials beat grants):
+ *
+ *   1. fetch KeyUser; if missing → undefined (no binding exists)
+ *   2. KeyUser.revokedAt set → false (subject-level ban beats everything)
+ *   3. manual-override layer (LIVE SigningConditions only):
+ *        - live matching per-(method,kind) deny → false
+ *        - live matching per-(method,kind) grant → true
+ *   4. live token grant: a redeemed Token bound to this KeyUser that is
+ *      neither revoked nor expired pairs the user (`connect`) outright and,
+ *      via its policy, governs signing. Token expiry/revoke are evaluated
+ *      HERE, every request — not photocopied at redeem (#24).
+ *   5. else → undefined (caller's requestPermission flow may prompt an admin)
+ *
+ * Unlike the pre-#25 algorithm, token grants are no longer materialized into
+ * SigningCondition rows at redeem (Backend.applyToken stopped photocopying),
+ * so step 4 is the live source of truth for token lifecycle. The override
+ * layer (step 3) is manual-only and now carries its own lifecycle, so an
+ * expired/revoked override stops granting too.
+ *
+ * Supersedes the #11 algorithm; closes the materialization-drift family
+ * behind #24. See aiolabs/nsecbunkerd#25.
+ */
 export async function checkIfPubkeyAllowed(
     keyName: string,
     remotePubkey: string,
     method: IMethod,
-    payload?: string | NostrEvent
+    payload?: string | NostrEvent,
 ): Promise<boolean | undefined> {
-    // find KeyUser
+    // One clock reading for the whole decision.
+    const now = new Date();
+
+    // Step 1: find KeyUser.
     const keyUser = await prisma.keyUser.findUnique({
         where: { unique_key_user: { keyName, userPubkey: remotePubkey } },
     });
@@ -16,59 +48,113 @@ export async function checkIfPubkeyAllowed(
         return undefined;
     }
 
-    // find SigningCondition
+    // Step 2: subject-level revoke (sticky ban, beats everything).
+    if (keyUser.revokedAt) {
+        return false;
+    }
+
+    const live = liveWhere(now);
+
+    // Step 3: live matching per-(method, kind) override — deny beats grant.
+    // (Subject-level "reject all from this user" is KeyUser.revokedAt, applied
+    // at step 2 via the revoke_user admin command. There is no method='*'
+    // SigningCondition sentinel — nothing writes one.)
     const signingConditionQuery = requestToSigningConditionQuery(method, payload);
 
-    const explicitReject = await prisma.signingCondition.findFirst({
+    const liveDeny = await prisma.signingCondition.findFirst({
-        where: {
+        where: { keyUserId: keyUser.id, ...signingConditionQuery, allowed: false, ...live },
-            keyUserId: keyUser.id,
-            method: '*',
-            allowed: false,
-        }
     });
 
-    if (explicitReject) {
+    if (liveDeny) {
-        console.log(`explicit reject`, explicitReject);
         return false;
     }
 
-    const signingCondition = await prisma.signingCondition.findFirst({
+    const liveGrant = await prisma.signingCondition.findFirst({
+        where: { keyUserId: keyUser.id, ...signingConditionQuery, allowed: true, ...live },
+    });
+
+    if (liveGrant) {
+        return true;
+    }
+
+    // Step 4: live token grant.
+    //
+    // A redeemed token that is live (not revoked, not past expiry) grants
+    // `connect` (the pairing) outright, and grants other methods when its
+    // policy has a matching PolicyRule. The live filter is what closes #24:
+    // an expired or revoked token simply stops matching here, every request,
+    // with no photocopy to outlive it.
+    if (method === 'connect') {
+        const liveToken = await prisma.token.findFirst({
+            where: { keyUserId: keyUser.id, ...live },
+        });
+
+        if (liveToken) {
+            return true;
+        }
+    } else {
+        // PolicyRule.kind matching:
+        //   - exact match against the stringified payload kind (matches the
+        //     create_new_policy.ts storage format `rule.kind.toString()`)
+        //   - 'all' literal matches any kind
+        //   - NULL kind is a defensive wildcard — no current writer emits a
+        //     null-kind rule, but treat it as a wildcard rather than failing
+        //     closed silently if one ever appears (raw SQL, future code).
+        const payloadKindString =
+            method === 'sign_event' && typeof payload === 'object' && payload?.kind !== undefined
+                ? payload.kind.toString()
+                : undefined;
+
+        const kindMatchers: Array<{ kind: string | null }> = [{ kind: null }, { kind: 'all' }];
+        if (payloadKindString !== undefined) {
+            kindMatchers.push({ kind: payloadKindString });
+        }
+
+        const policyAllowance = await prisma.token.findFirst({
             where: {
                 keyUserId: keyUser.id,
-            ...signingConditionQuery,
+                ...live,
-        }
+                policy: {
+                    rules: {
+                        some: {
+                            method,
+                            OR: kindMatchers,
+                        },
+                    },
+                },
+            },
         });
 
-    // if no SigningCondition found, return undefined
+        if (policyAllowance) {
-    if (!signingCondition) {
+            return true;
+        }
+    }
+
+    // Step 5: no live override and no live token grant matched. Caller's
+    // requestPermission flow may still prompt the admin out-of-band.
     return undefined;
 }
 
-    const allowed = signingCondition.allowed;
+/**
-
+ * Sign-time auth method names follow the NIP-46 wire convention as
-    // Check if the key user has been revoked
+ * NDK 3.x's `NDKNip46Backend` passes them through to `pubkeyAllowed`
-    if (allowed) {
+ * verbatim (it stopped normalizing `nip04_encrypt`/`nip04_decrypt`
-        const revoked = await prisma.keyUser.findFirst({
+ * to `encrypt`/`decrypt` somewhere between 2.8.1 and current
-            where: {
+ * upstream).
-                id: keyUser.id,
+ *
-                revokedAt: { not: null },
+ * lnbits's `_ensure_policy` writes `PolicyRule.method` using the same
-            }
+ * wire-name vocabulary (`nip04_encrypt`, `nip04_decrypt`,
-        });
+ * `nip44_encrypt`, `nip44_decrypt`, `sign_event`, `get_public_key`,
-
+ * `connect`, `ping`). With the wire-name vocabulary on both sides,
-        if (revoked) {
+ * the post-#11 live-policy join (step 4 of `checkIfPubkeyAllowed`)
-            return false;
+ * naturally matches lnbits's stored rules — no `encrypt → nip04_encrypt`
-        }
+ * adapter layer needed.
-    }
+ *
-
+ * Source the type from NDK itself so it can't drift across future
-    if (allowed === true || allowed === false) {
+ * NDK bumps; if NDK adds a new method (e.g. `nip60_*`) we pick it up
-        console.log(`found signing condition`, signingCondition);
+ * for free.
-        return allowed;
+ */
-    }
+export type IMethod = NIP46Method;
-
-    return undefined;
-}
-
-export type IMethod = "connect" | "sign_event" | "encrypt" | "decrypt" | "ping";
 
 export type IAllowScope = {
     kind?: number | 'all';
@@ -78,10 +164,14 @@ export function requestToSigningConditionQuery(method: IMethod, payload?: string
     const signingConditionQuery: any = { method };
 
     switch (method) {
-        case 'sign_event':
+        case 'sign_event': {
-            signingConditionQuery.kind = { in: [ payload?.kind?.toString(), 'all' ] };
+            const kindString = (typeof payload === 'object' && payload?.kind !== undefined)
+                ? payload.kind.toString()
+                : undefined;
+            signingConditionQuery.kind = { in: [kindString, 'all'] };
             break;
         }
+    }
 
     return signingConditionQuery;
 }
@@ -126,20 +216,3 @@ export async function allowAllRequestsFromKey(
         console.log('allowAllRequestsFromKey', e);
     }
 }
-
-export async function rejectAllRequestsFromKey(remotePubkey: string, keyName: string): Promise<void> {
-    // Upsert the KeyUser with the given remotePubkey
-    const upsertedUser = await prisma.keyUser.upsert({
-        where: { unique_key_user: { keyName, userPubkey: remotePubkey } },
-        update: { },
-        create: { keyName, userPubkey: remotePubkey },
-    });
-
-    // Create a new SigningCondition for the given KeyUser and set allowed to false
-    await prisma.signingCondition.create({
-        data: {
-            allowed: false,
-            keyUserId: upsertedUser.id,
-        },
-    });
-}

src/daemon/lib/acl/lifecycle.ts

@@ -0,0 +1,40 @@
+/**
+ * Pure grant-lifecycle logic, extracted from the ACL so it can be unit-tested
+ * without a database and reused verbatim at redeem time and sign time.
+ *
+ * The original #24 bug was possible because redeem-time checked expiry and
+ * sign-time didn't — two definitions of "valid" that drifted. Defining "is
+ * this grant valid right now?" exactly once makes them impossible to disagree.
+ * See aiolabs/nsecbunkerd#25.
+ */
+
+/** The lifecycle fields every grant (Token, SigningCondition) carries. */
+export type Lifecycle = {
+    revokedAt?: Date | null;
+    expiresAt?: Date | null;
+};
+
+/**
+ * "Is this grant valid right now?" — the single lifecycle predicate. A grant
+ * is live iff it has not been revoked and its expiry (if any) is still in the
+ * future. Expiry is treated as exclusive at the boundary: a grant whose
+ * `expiresAt` equals `now` is already dead.
+ */
+export function grantIsLive(grant: Lifecycle, now: Date = new Date()): boolean {
+    if (grant.revokedAt) return false;
+    if (grant.expiresAt && grant.expiresAt.getTime() <= now.getTime()) return false;
+    return true;
+}
+
+/**
+ * `grantIsLive` expressed as a Prisma `where` fragment, so the live filter
+ * runs in the query rather than in app code after the fetch. `now` is threaded
+ * in explicitly so a single request evaluates every row against one clock
+ * reading. Kept in lockstep with `grantIsLive` (see lifecycle.test.ts).
+ */
+export function liveWhere(now: Date) {
+    return {
+        revokedAt: null,
+        OR: [{ expiresAt: null }, { expiresAt: { gt: now } }],
+    };
+}

src/daemon/lib/relay-reconnect.ts

@@ -0,0 +1,101 @@
+import NDK from "@nostr-dev-kit/ndk";
+
+/**
+ * Attaches an aggressive-reconnect supervisor to an NDK instance.
+ *
+ * NDK 3.x's per-relay connectivity state machine gives up retrying after
+ * a few consecutive fast-fail (e.g. ECONNREFUSED returns in <1 ms)
+ * connection attempts:
+ *
+ *   1. Each attempt's duration is recorded in `_connectionStats.durations`.
+ *   2. After every 3 attempts, `isFlapping()` checks the std-dev of those
+ *      durations against `FLAPPING_THRESHOLD_MS` (1 second). Three fast
+ *      failures look identical → tiny std-dev → flapping=true → status
+ *      transitions to FLAPPING and the per-relay retry stops.
+ *   3. `NDKPool.handleFlapping` catches the event and reschedules a
+ *      reconnect via doubling backoff (5s → 10s → 20s → 40s → 80s …),
+ *      growing unbounded.
+ *
+ * For nsecbunkerd, where the admin relay is typically a single relay we
+ * **must** stay subscribed to, "disconnected for 80+s after every dev
+ * restart" is the failure mode users hit (aiolabs/nsecbunkerd#20). The
+ * pool's doubling backoff is too pessimistic for our use case.
+ *
+ * This helper sidesteps the give-up path: when the pool emits `flapping`
+ * (the symptom that NDK has internally given up), or when we see the
+ * relay disconnect outside our own request, we manually call
+ * `relay.connect()` with a SHORT capped delay. Successful connect resets
+ * the attempt counter so a future disconnect storm doesn't grow the
+ * delay.
+ *
+ * Trade-off: we may hammer a permanently-down relay every 10s. That's
+ * fine for a bunker — being disconnected silently is strictly worse than
+ * a retry storm against localhost. Acceptable because:
+ *   - The bunker's primary relay is typically on the same host or LAN
+ *     (`ws://lnbits:5001/...`); TCP RSTs are cheap.
+ *   - Public-relay setups can layer external supervision on top if they
+ *     care about retry pressure.
+ */
+export function attachIndefiniteReconnect(ndk: NDK, label: string): void {
+    const RECONNECT_BASE_MS = 1_000;
+    const RECONNECT_CAP_MS = 10_000;
+
+    const attempts = new Map<string, number>();
+    const pending = new Map<string, NodeJS.Timeout>();
+
+    const reconnectDelay = (n: number): number =>
+        Math.min(RECONNECT_BASE_MS * 2 ** n, RECONNECT_CAP_MS);
+
+    const scheduleReconnect = (relay: any): void => {
+        const url: string = relay.url;
+        if (pending.has(url)) return;
+        const n = attempts.get(url) ?? 0;
+        const delay = reconnectDelay(n);
+        console.log(
+            `🔁 ${label}: scheduling reconnect to ${url} in ${delay}ms ` +
+            `(attempt ${n + 1}, overriding NDK give-up)`
+        );
+        const timer = setTimeout(() => {
+            pending.delete(url);
+            attempts.set(url, n + 1);
+            relay.connect().catch((e: any) => {
+                console.log(
+                    `❌ ${label}: manual reconnect to ${url} failed: ` +
+                    `${e?.message ?? e}`
+                );
+                // Don't recurse here — the next 'flapping' or 'disconnect'
+                // event will fire and schedule another attempt.
+            });
+        }, delay);
+        pending.set(url, timer);
+    };
+
+    ndk.pool.on("flapping", (relay: any) => {
+        console.log(
+            `⚠️  ${label}: NDK flagged ${relay.url} as flapping ` +
+            `(connectivity machine gave up internally)`
+        );
+        scheduleReconnect(relay);
+    });
+
+    ndk.pool.on("relay:disconnect", (relay: any) => {
+        scheduleReconnect(relay);
+    });
+
+    ndk.pool.on("relay:connect", (relay: any) => {
+        const url: string = relay.url;
+        const n = attempts.get(url) ?? 0;
+        if (n > 0) {
+            console.log(
+                `✅ ${label}: recovered ${url} after ${n} manual reconnect ` +
+                `attempt(s)`
+            );
+        }
+        attempts.delete(url);
+        const timer = pending.get(url);
+        if (timer) {
+            clearTimeout(timer);
+            pending.delete(url);
+        }
+    });
+}

src/daemon/run.ts

@@ -1,10 +1,7 @@
 import NDK, { NDKPrivateKeySigner, Nip46PermitCallback, Nip46PermitCallbackParams } from '@nostr-dev-kit/ndk';
-import { nip19 } from 'nostr-tools';
+import { nip19, utils as nostrUtils } from 'nostr-tools';
 import { Backend } from './backend/index.js';
-import {
+import { checkIfPubkeyAllowed } from './lib/acl/index.js';
-    IMethod,
-    checkIfPubkeyAllowed,
-} from './lib/acl/index.js';
 import AdminInterface from './admin/index.js';
 import { IConfig } from '../config/index.js';
 import { NDKRpcRequest } from '@nostr-dev-kit/ndk';
@@ -18,6 +15,7 @@ import FastifyView from '@fastify/view';
 import Handlebars from "handlebars";
 import {authorizeRequestWebHandler, processRequestWebHandler} from "./web/authorize.js";
 import {processRegistrationWebHandler} from "./web/authorize.js";
+import { attachIndefiniteReconnect } from "./lib/relay-reconnect.js";
 
 export type Key = {
     name: string;
@@ -38,8 +36,7 @@ function getKeys(config: DaemonConfig) {
         const keys: Key[] = [];
 
         for (const [name, nsec] of Object.entries(config.keys)) {
-            const hexpk = nip19.decode(nsec).data as string;
+            const user = await new NDKPrivateKeySigner(nsec).user();
-            const user = await new NDKPrivateKeySigner(hexpk).user();
             const key = {
                 name,
                 npub: user.npub,
@@ -108,7 +105,7 @@ function signingAuthorizationCallback(keyName: string, adminInterface: AdminInte
         }
 
         try {
-            const keyAllowed = await checkIfPubkeyAllowed(keyName, remotePubkey, method as IMethod, payload);
+            const keyAllowed = await checkIfPubkeyAllowed(keyName, remotePubkey, method, payload);
 
             if (keyAllowed === true || keyAllowed === false) {
                 console.log(`🔎 ${nip19.npubEncode(remotePubkey)} is ${keyAllowed ? 'allowed' : 'denied'} to ${method} with key ${keyName}`);
@@ -164,11 +161,17 @@ class Daemon {
             explicitRelayUrls: config.nostr.relays,
         });
         this.ndk.pool.on('relay:connect', (r) => console.log(`✅ Connected to ${r.url}`) );
-        this.ndk.pool.on('relay:notice', (n, r) => { console.log(`👀 Notice from ${r.url}`, n); });
+        this.ndk.pool.on('notice', (r, n) => { console.log(`👀 Notice from ${r.url}`, n); });
 
         this.ndk.pool.on('relay:disconnect', (r) => {
             console.log(`🚫 Disconnected from ${r.url}`);
         });
+
+        // Override NDK's "give up after detecting flapping" behavior so the
+        // bunker's backend NDK keeps trying to reconnect indefinitely.
+        // Without this, an ECONNREFUSED storm at boot (relay not yet up)
+        // permanently strands the bunker. See aiolabs/nsecbunkerd#20.
+        attachIndefiniteReconnect(this.ndk, 'backend');
     }
 
     async startWebAuth() {
@@ -206,9 +209,129 @@ class Daemon {
                 continue;
             }
 
-            const nsec = nip19.nsecEncode(settings.key);
+            // nostr-tools v2: `nsecEncode` takes `Uint8Array`, not hex string.
+            // pragma: allowlist secret
+            // `settings.key` is the hex-encoded private key from config.
+            const nsec = nip19.nsecEncode(nostrUtils.hexToBytes(settings.key));
             this.loadNsec(keyName, nsec);
         }
+
+        // Boot-time autounlock of encrypted-at-rest keys. Off by default;
+        // enabled by setting NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE or
+        // NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE_FILE. See docs/AUTOUNLOCK.md
+        // for the security trade-off and aiolabs/nsecbunkerd#16 for the
+        // design rationale.
+        await this.maybeAutounlock();
+    }
+
+    /**
+     * Boot-time autounlock for encrypted keys.
+     *
+     * Reads a passphrase from one of two mutually exclusive env vars:
+     *   - NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE       — literal passphrase
+     *   - NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE_FILE  — path to a file containing
+     *                                                the passphrase (newline-trimmed)
+     *
+     * If neither is set, this is a no-op — the deployment opted out and
+     * keys remain locked until an admin `unlock_key` RPC fires per key
+     * per restart (today's default).
+     *
+     * If both are set, throws at boot — ambiguous config.
+     *
+     * Otherwise: enumerates `Key` table rows where `deletedAt IS NULL`,
+     * calls `unlockKey(keyName, passphrase)` per row. Sequential, with
+     * continue-on-error so one bad row doesn't block the rest of the
+     * fleet. Per-key INFO/WARN/ERROR log + one summary line at the end.
+     *
+     * `unlockKey` is idempotent post-#16 — calling it against a key that
+     * was already loaded via the unencrypted paths above is safe (returns
+     * true without spawning a duplicate Backend).
+     *
+     * Single-passphrase invariant: every `create_new_key(name, passphrase)`
+     * uses the same passphrase in our usage today, so one autounlock
+     * passphrase covers every encrypted key. Per-key passphrase support
+     * is a separate feature (out of scope — see issue #16).
+     */
+    async maybeAutounlock(): Promise<void> {
+        const literal = process.env.NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE;
+        const filePath = process.env.NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE_FILE;
+
+        if (literal && filePath) {
+            throw new Error(
+                'Autounlock: NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE and ' +
+                'NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE_FILE are mutually exclusive. ' +
+                'Set exactly one (or neither, to leave autounlock off).'
+            );
+        }
+
+        if (!literal && !filePath) {
+            return; // autounlock off (default)
+        }
+
+        let passphrase: string;
+        let source: string;
+        if (literal) {
+            passphrase = literal;
+            source = 'NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE';
+        } else {
+            const fs = await import('fs');
+            try {
+                passphrase = fs.readFileSync(filePath!, 'utf8').replace(/\r?\n$/, '');
+            } catch (e: any) {
+                throw new Error(
+                    `Autounlock: failed to read passphrase file ${filePath}: ${e.message}`
+                );
+            }
+            source = `NSEC_BUNKER_AUTOUNLOCK_PASSPHRASE_FILE=${filePath}`;
+        }
+
+        // Enumerate encrypted-at-rest keys from `config.allKeys`. The
+        // Prisma `Key` table is only populated by the NIP-05 `create_account`
+        // path (which stores keys plain-at-rest in nsecbunker.json);
+        // `create_new_key` provisions keys with the `{iv, data}` encrypted
+        // shape directly into the JSON blob without a Prisma row. So the
+        // canonical "what's encrypted at rest" source is `allKeys` filtered
+        // to entries carrying `iv`+`data` — that's the set of keys for
+        // which the manual `unlock_key` admin RPC was previously required
+        // per restart, and exactly the set we want to autounlock here.
+        //
+        // Plain-key entries (`{key: "..."}` shape, populated by `create_account`)
+        // were already loaded by the second loop in `startKeys` above and
+        // appear in `activeKeys` — `unlockKey`'s idempotency guard makes
+        // re-calling them safe but unnecessary, so we filter them out for
+        // log clarity.
+        const candidates = Object.entries(this.config.allKeys || {})
+            .filter(([, entry]) =>
+                entry && typeof entry === 'object' && 'iv' in entry && 'data' in entry
+            )
+            .map(([keyName]) => keyName);
+
+        const start = Date.now();
+        let success = 0;
+
+        for (const keyName of candidates) {
+            try {
+                const ok = await this.unlockKey(keyName, passphrase);
+                if (ok) {
+                    console.log(`🔓 autounlock: unlocked ${keyName}`);
+                    success++;
+                } else {
+                    console.warn(
+                        `⚠️  autounlock: unlockKey returned false for ${keyName} ` +
+                        `(likely wrong passphrase — encrypted under a different secret?)`
+                    );
+                }
+            } catch (e: any) {
+                console.error(
+                    `❌ autounlock: ${keyName} failed: ${e?.message ?? e}`
+                );
+            }
+        }
+
+        const elapsed = Date.now() - start;
+        console.log(
+            `🔓 autounlock: enabled (source=${source}), unlocked ${success}/${candidates.length} keys in ${elapsed}ms`
+        );
     }
 
     async start() {
@@ -226,39 +349,44 @@ class Daemon {
      */
     async startKey(name: string, nsec: string) {
         const cb = signingAuthorizationCallback(name, this.adminInterface);
-        let hexpk: string;
+        // NDK 3.x's `NDKPrivateKeySigner` accepts nsec1 or hex directly
-
+        // (see `core/src/signers/private-key/index.ts` `@ai-guardrail`
-        if (nsec.startsWith('nsec1')) {
+        // — "DO NOT use nip19.decode() to convert nsec to hex before
-            try {
+        // passing it here"). The bech32-decode workaround for #8 was
-                const key = new NDKPrivateKeySigner(nsec);
+        // tied to NDK 2.8.1's old constructor behavior and is no
-                hexpk = key.privateKey!;
+        // longer needed post-#14 NDK bump.
-            } catch(e) {
+        const backend = new Backend(this.ndk, this.fastify, nsec, cb, this.config.baseUrl);
-                console.error(`Error loading key ${name}:`, e);
-                return
-            }
-        } else {
-            hexpk = nsec;
-        }
-        
-        const backend = new Backend(this.ndk, this.fastify, hexpk, cb, this.config.baseUrl);
         await backend.start();
     }
 
     async unlockKey(keyName: string, passphrase: string): Promise<boolean> {
+        // Idempotency guard: if a Backend instance already exists for this
+        // keyName, the key is already unlocked and the relay subscription
+        // for its kind-24133 channel is already active. Calling startKey
+        // again would spawn a SECOND Backend with a duplicate subscription
+        // — wire events would be handled twice, with race/amplification
+        // hazards on the response side. Return success without re-running
+        // startKey so callers (admin `unlock_key` RPC, autounlock loop,
+        // belt-and-suspenders fallback paths) can fire safely against
+        // already-unlocked keys.
+        if (this.activeKeys[keyName]) {
+            return true;
+        }
+
         const keyData = this.config.allKeys[keyName];
         const { iv, data } = keyData;
 
         const nsec = decryptNsec(iv, data, passphrase);
         this.activeKeys[keyName] = nsec;
 
-        this.startKey(keyName, nsec);
+        await this.startKey(keyName, nsec);
 
         return true;
     }
 
-    loadNsec(keyName: string, nsec: string) {
+    async loadNsec(keyName: string, nsec: string) {
         this.activeKeys[keyName] = nsec;
 
-        this.startKey(keyName, nsec);
+        await this.startKey(keyName, nsec);
     }
 }

tests/lifecycle.test.ts

@@ -0,0 +1,44 @@
+import { test } from 'node:test';
+import assert from 'node:assert/strict';
+import { grantIsLive, liveWhere } from '../src/daemon/lib/acl/lifecycle';
+
+// Fixed reference clock so the assertions don't depend on wall time.
+const now = new Date('2026-06-19T12:00:00.000Z');
+const past = new Date(now.getTime() - 60_000);
+const future = new Date(now.getTime() + 60_000);
+
+test('grantIsLive: no revoke, no expiry -> live', () => {
+    assert.equal(grantIsLive({}, now), true);
+    assert.equal(grantIsLive({ revokedAt: null, expiresAt: null }, now), true);
+});
+
+test('grantIsLive: future expiry -> live', () => {
+    assert.equal(grantIsLive({ expiresAt: future }, now), true);
+});
+
+test('grantIsLive: past expiry -> dead (the #24 case the old code missed at sign time)', () => {
+    assert.equal(grantIsLive({ expiresAt: past }, now), false);
+});
+
+test('grantIsLive: expiry exactly now -> dead (boundary is exclusive)', () => {
+    assert.equal(grantIsLive({ expiresAt: new Date(now.getTime()) }, now), false);
+});
+
+test('grantIsLive: revoked -> dead even with a future expiry (revoke wins)', () => {
+    assert.equal(grantIsLive({ revokedAt: past, expiresAt: future }, now), false);
+});
+
+test('grantIsLive: defaults now to the current time', () => {
+    assert.equal(grantIsLive({ expiresAt: new Date(Date.now() + 3_600_000) }), true);
+    assert.equal(grantIsLive({ expiresAt: new Date(Date.now() - 3_600_000) }), false);
+});
+
+// liveWhere is the SQL mirror of grantIsLive; pin its shape so the two
+// can't silently drift (a drift would re-open the redeem-vs-sign gap #25
+// exists to close).
+test('liveWhere: mirrors grantIsLive as a prisma where-fragment', () => {
+    assert.deepEqual(liveWhere(now), {
+        revokedAt: null,
+        OR: [{ expiresAt: null }, { expiresAt: { gt: now } }],
+    });
+});