@forgesworn/kenspeckle

Verified relationships across three social distances — kin (family), kith (mutually verified), ken (one-way recognised) — with the handshake, bond ceremony, ken trust-store, local presence discovery, and invite flows. Protocol-neutral, pure functions, no storage, no graph traversal.

An app wants to hold a user's relationships — family, people they've met and verified, public keys they recognise — without becoming a social network: no feed, no follower count, no central social graph, no shadow profiles. It also wants two ceremonies most contact systems lack: a bond (two people mutually verify each other out-of-band by speaking short rotating words from a shared secret) and a live key-control challenge (a public figure proves an in-game account really holds the key you pinned).

kenspeckle is those primitives as plain data + pure functions. It does not store anything, render anything, or traverse a graph (those are explicit non-goals). It holds relationships, derives ceremony words, pins and challenges recognised keys, and tests a user's own contacts against a server's published presence filter (via the sibling @forgesworn/tessera-kit) — locally, never uploading contacts.

Naming. kenspeckle (Scots: easily recognised, known by sight) was developed as @forgesworn/kindred and renamed before first publish — "Kindred" is now the ForgeSworn app-suite brand. The wire protocol keeps the historical kindred naming, frozen: 'kindred:bond', 'kindred:members:', 'kindred-bond', kinds 30444/30445, and the KINDRED_* / KindredEntry / KindredTier identifiers are unchanged — renaming them would change every deployed bond word and published filter.

Honest scope (read SECURITY.md): persona-scoping is a consumer obligation kenspeckle assists but cannot guarantee (no storage). NIP-05 is a DNS/TLS/HTTP TOFU anchor, not key-continuity. attributeSignature alone is replayable — impersonation resistance needs verifyKeyControl against a fresh nonce. Private annotations are never serialised. A forged presence filter is a doxxing primitive — verify it against a pinned server key.

npm i @forgesworn/kenspeckle

ESM-only, Node ≥ 22. nostr-tools is a peer dependency (so you and kenspeckle share one event type) — install it alongside.

kenspeckle depends on @forgesworn/tessera-kit (^0.2.1, from npm), which provides the signed membership filters that discovery verifies. A scripts/check-publishable-deps.mjs guard runs in prepack and refuses to publish while any dependencies / optionalDependencies / peerDependencies entry is a git/file/http(s) spec, so kenspeckle can only ship against registry versions.

Two people exchange a handshake, derive the byte-exact shared secret, then read each other short rotating words. The word I speak differs from the word I expect (directional — the listener can't parrot it back).

import { buildHandshakePayload, parseHandshakePayload } from '@forgesworn/kenspeckle/handshake'
import { deriveBondSecret, bondWords, verifyBondWord } from '@forgesworn/kenspeckle/bond'

// (1) I present my persona + a fresh nonce over QR/NFC/relay; they parse it.
const blob = buildHandshakePayload({ pubkey: myPubHex, nonce: my16ByteHexNonce })
const theirs = parseHandshakePayload(receivedBytes) // validates; displayName is VERBATIM

// (2) Each side derives the SAME ECDH secret. (myPriv × their x-only pubkey, hash x.)
const secret = deriveBondSecret(myPrivHex, theirs.pubkey)

// (3) Read words for an agreed counter. I say `mine`; I expect to hear `theirs`.
const { mine } = bondWords(secret, myPubHex, theirs.pubkey, counter)
speakAloud(mine)

// (4) Verify what THEY spoke (constant-time compare under the hood).
const result = verifyBondWord(secret, myPubHex, theirs.pubkey, counter, whatTheySaid)
if (result.ok) { /* bonded — persist a KithEntry, encrypt sharedSecret at rest */ }

Migrating from signet-app's signet-me? By DEFAULT bondWords uses namespace 'kindred:bond', so its words differ from signet-me's ('signet:me') — a naive migration changes the words. To cross-verify with a peer who hasn't migrated yet, reproduce signet-me's words by passing bondWords(secret, myPub, theirPub, counter, { namespace: 'signet:me' }) and verifyBondWord(…, { namespace: 'signet:me', tolerance: 1 }) (each seat passes its OWN pubkey first). Both peers must either upgrade together or pass this signet-me namespace.

attributeSignature ("did the pinned key sign this old event?") is replayable. To bind an in-game account to the real key, issue a fresh nonce and verify the claimant signs it.

import { pinKenFromNip05, buildKeyControlChallenge, verifyKeyControl } from '@forgesworn/kenspeckle/ken'

// Pin via NIP-05 — refuses to pin unless the name resolves (TOFU; HTTPS-only fetch).
const entry = await pinKenFromNip05('mrbeast@example.com', myGamingPersonaPubHex, fetch)

// Live challenge: the claimant signs an event whose `content` IS the nonce.
const { nonce } = buildKeyControlChallenge()
const signed = await askClaimantToSign(nonce) // their client builds + signs it
const proof = verifyKeyControl(entry, nonce, signed)
if (proof.ok) { /* the account in front of me holds the pinned key, live (no replay) */ }

A server publishes a signed, non-enumerable presence filter (@forgesworn/tessera-kit). The client parses it, verifies it against a pinned server key, then tests its own contacts — scoped to one persona.

// `parseFilter` is re-exported from `./discovery` (it delegates to `@forgesworn/tessera-kit`
// internally) so a consumer importing only kenspeckle's discovery subpath never has to reach past
// it into tessera-kit directly.
import { discoverPresent, parseFilter, parseFilterPublication } from '@forgesworn/kenspeckle/discovery'

// Pull the kind-30444 publication for the (namespace, serverId) YOU chose to fetch — as of 0.2.0
// this REQUIRES opts, since @forgesworn/tessera-kit now binds the in-blob signature to a `context`
// string built ONLY from opts (kindred: `kindred:members:<namespace>:<serverId>`), never from the
// event's own d-tag — that is what stops a relay from relabelling a substituted blob. Verifies BOTH
// the Nostr event sig and the in-blob Schnorr sig (bound to that context), that the event's own
// d-tag equals it, and — default true — that the event's signer IS the in-blob signer
// (`requireAuthorIsSigner`). Returns null on any failure.
const pub = parseFilterPublication(rawWireEvent, {
  namespace: 'com.example.game',
  serverId: 'play.example.com',
}) // pass the RAW event — not a spread-mutated one
if (!pub) throw new Error('untrusted publication')

// Provenance gate: the IN-BLOB signer must be a key you pinned out-of-band.
if (pub.signerPubkeyHex !== PINNED_SERVER_PUBKEY) throw new Error('unpinned server')

// Test MY contacts, scoped to one persona (throws on mixed-owner input — anti-correlation; also
// throws if a salt is passed to an OPEN filter, or omitted for a KEYED one — either mismatch
// silently matches nothing otherwise).
const filter = parseFilter(pub.blob)
const present = discoverPresent(filter, myEntries, myGamingPersonaPubHex /*, salt if keyed */)
// `present` is a CANDIDATE list — confirm-on-connect with a ken key-control challenge before acting.

The . entry carries the model + local ops; the protocol ceremonies are separate subpaths (so a consumer that only needs ./bond doesn't pull the discovery / @forgesworn/tessera-kit graph).

Pure wire ownership shared by Signet producers and companion consumers: buildPairingUri / parsePairingRequest, buildPairingAck / parsePairingAck, applyCompanionSnapshot, the pairing/snapshot types, and the frozen ACK_KIND, SNAPSHOT_KIND, and SNAPSHOT_D_TAG constants. Relay subscriptions, encryption, device keys, clocks, storage, retry policy and UI remain in the consuming app.

Return rail — WireKen, buildReturnEnvelope / parseReturnEnvelope, landReturnedKen, and the RETURN_D_TAG / RETURN_ADDITIONS_CAP / RETURN_CORROBORATIONS_CAP constants. A companion app holds no identity keys, so it can never mint a kith/kin bond — it can only propose a ken. It can, however, carry the provenance it claims (claimedProvenance / claimedCorroborations): the companion is often where the evidence actually is, and scanning someone's code while standing next to them is real in-person provenance. landReturnedKen preserves that claim as a corroboration namespaced companion:<appName>:<locator> (with confirmedAt clamped to now), while the primary provenance stays { source:'manual', locator:'companion:<appName>' } — what Signet can attest itself. A claim is preserved, never promoted: no app but Signet is a source of identity truth.

buildHandshakePayload(p) → bytes; parseHandshakePayload(blob) (8192-byte cap, all fields validated, displayName returned verbatim). Type HandshakePayload.

deriveBondSecret(myPrivHex, theirPubHex) (byte-exact ECDH); bondWords(secret, aPub, bPub, counter, opts?) → { mine, theirs }; verifyBondWord(…, spoken, opts?) → { ok }; buildBondAttestation({ subjectPubHex, summary? }) → kind-31000 EventTemplate (type:'kindred-bond').

Retracting a bond: buildBondRevocation({ subjectPubHex, reason? }) is now the primary way to retract — it emits a kind-31000 ["status","revoked"] event (nostr-attestations createRevocation) at the SAME addressable slot, so it replaces the attestation and verifyBondAttestation rejects it as revoked. Sign it with the key that signed the attestation. retractBondAssertion(assertion, { attesterPubHex, subjectPubHex }) remains as a supplement: a kind-5 NIP-09 deletion carrying e (the attestation event id), a (its addressable coordinate) and k (kind 31000) tags. Const KINDRED_BOND_NAMESPACE. The optional opts ({ namespace? }, plus tolerance? on verifyBondWord) exist for signet-me migration compatibility — { namespace: 'signet:me', tolerance: N } reproduces signet-app's signet-me words so a migrated contact can cross-verify with an un-migrated peer; defaults ('kindred:bond' + tolerance 0) keep the existing behaviour but differ from signet-me's words.

deriveCeremonyCounter(nonceAHex, nonceBHex) → number — the counter for an in-person bond ceremony: both parties derive the SAME counter from their two handshake nonces (SHA-256, domain-tagged CEREMONY_COUNTER_TAG), symmetric in argument order. timeCounter(nowSec, periodSec) → number — Math.floor(nowSec / periodSec), for a later re-verification (no fresh nonces on hand); the result must fit the uint32 counter range, otherwise it throws (it is not clamped). A fixed counter is not recommended — see PROTOCOL.md §2. normalizeSpokenWord(s) → string — NFKC + trim + lowercase; verifyBondWord already applies this to spoken before comparing, so callers don't need to.

import { deriveCeremonyCounter } from '@forgesworn/kenspeckle/bond'

// Both handshake nonces are exchanged already (see the kith example above).
const counter = deriveCeremonyCounter(myNonceHex, theirs.nonce) // same value on both sides
const { mine } = bondWords(secret, myPubHex, theirs.pubkey, counter)

pinKen(p) / pinKenFromNip05(nip05, ownerPubkeyHex, fetch) (refuse-on-mismatch TOFU); buildKeyControlChallenge(); verifyKeyControl(entry, nonce, signedEvent, opts?) (live, fail-closed; opts.expectedCreatedAt/maxAgeSec/verifierTag are an OPT-IN, backward-compatible partial mitigation for a verifier-relay attack — see the doc comment); attributeSignature(entry, event) (replayable — credit, don't authenticate); resolveKen (propose-not-flip; a NIP-05 resolving to a key already in previousPubkeys is flagged rotation.rollback: true rather than proposed as an ordinary rotation) / acceptKenRotation(entry, opts?) (throws if already accepted, if newPubkey equals the current pin, or — unless opts: { allowRevert: true } — if the rotation is a flagged rollback) / revokeKen / dropKen (consumer deletes).

Both pinKen and addCorroboration(entry, provenance) run the SAME provenance validation, the same MAX_CORROBORATIONS (64) cap, and reject the reserved companion: locator prefix — a builder can no longer mint an entry the parser would then refuse.

addCorroboration(entry, provenance) records another independent channel agreeing this key is this person — pure, appends in observation order, never touches the primary provenance, and deliberately does not de-duplicate (a re-check a year later is new recency evidence). summarizeKenProvenance(entry) reports { confirmations, claimed, distinctSources, distinctLocators, sources, mostRecentAt, oldestAt }.

NIP-05 handling. Every entry point that can fetch a nip05 (pinKen, pinKenFromNip05, resolveKen) validates a STRICT local@domain shape — no port, no userinfo, no path/query/fragment, and a last label that is alphabetic or xn--, so no IP literal in any form (127.1, 0x7f.1 included). The parser keeps a stored nip05 that 0.1.x accepted; resolveKen treats one that fails the strict check as unresolvable and never fetches it. The same check is exported as validateNip05(value) from ./ken (returns the value, throws on a bad one), so an app can pre-validate input and show its own message. The fetch is hardened: redirect:'error' (NIP-05 requires ignoring redirects), a bounded timeout, a size-capped body read before JSON.parse, and both halves lowercased before querying/looking up (NIP-05 names are case-insensitive).

claimed is the one to render alongside any total. A ken landed from a companion app can show six confirmations across six distinct sources with nothing verified first-hand — every record is a relayed claim. claimed counts the records in the reserved companion: locator namespace, so confirmations - claimed is what was actually confirmed first-hand. Show a total without it and you present relayed claims as verification.

It returns facts, not a trust score — on purpose. A scalar "strength" invites false precision and invites being used as an authorization input (if (strength > 0.7) allow), which is a security decision this primitive cannot underwrite; and weighting channels against each other (is one in-person worth two nip05?) is consumer policy, not a property of the data. It is also honest about its limits: distinctSources can understate independence (two different websites are both web) and distinctLocators can overstate it (one operator can serve two locators; DNS and the site behind it often share one point of compromise). Signals, not guarantees.

discoverPresent(filter, entries, ownerPubkey, saltHex?) (throws on mixed persona; also throws if saltHex is given for an OPEN filter or omitted for a KEYED one, or if saltHex is the empty string for a KEYED one (@forgesworn/tessera-kit 0.2.0 rejects an empty salt outright) — any of these would otherwise silently match nothing, a false "no friends here"); disclosureFor(filter: MembershipFilter) → DiscoveryDisclosure (pass the filter parseFilter returned); filterSignatureContext(namespace, serverId) → the kindred convention's filter-signature context string (@forgesworn/tessera-kit 0.2.0), identical to the d-tag value kindred:members:<namespace>:<serverId> — build it from the (namespace, serverId) you chose to fetch, never from a received event's own d-tag; buildFilterPublication(p) — p.blob must already be signed (signFilterBlob) for filterSignatureContext(p.namespace, p.serverId); this function verifies that itself and throws a clear error if it doesn't verify, so a blob signed for the wrong deployment is caught before publishing / parseFilterPublication(event, opts) — opts is now REQUIRED: { namespace, serverId, minEpoch?, requireAuthorIsSigner? } (BREAKING as of 0.2.0; previously optional, with namespace/serverId read off the event's own d-tag). Verifies the Nostr event sig, that the event's d-tag equals filterSignatureContext(opts.namespace, opts.serverId), the in-blob Schnorr sig bound to that same context (@forgesworn/tessera-kit 0.2.0 — closes cross-server/namespace filter substitution cryptographically), and — opts.requireAuthorIsSigner, default true — that the event's signer IS the in-blob signer; pass { requireAuthorIsSigner: false } to opt out for a different trust model. Returns null, never throws — including when opts is missing or opts.namespace/opts.serverId is invalid; parseFilterPublicationResult(event, opts) → { ok: true, value } | { ok: false, reason: FilterPublicationRejection } — the same checks as parseFilterPublication, but naming WHICH one failed instead of collapsing to null (parseFilterPublication is now a thin wrapper over this); aggregatorQuery(namespace); buildOptOutRequest(p, memberPrivHex). Re-exports parseFilter from @forgesworn/tessera-kit. Consts KINDRED_FILTER_KIND = 30444, KINDRED_OPTOUT_KIND = 30445.

buildJoinInvite(p, inviterPrivHex) → JoinInvite object; parseJoinInvite(blob, now?) (verifies sig + expiry); verifyBondAttestation(event, now?) → { ok, reason?, attesterPubHex?, subjectPubHex? } (single-attestation only — no counting/graph; fails with a reason for a revoked, expired, not-yet-active, or self-attestation). Type JoinInvite.

Invites are v2 only. Every JoinInvite carries v: 2. The signed digest is sha256(JSON.stringify(['kenspeckle-invite', 2, namespace, serverId, inviterPubkey, nonce, expiresAt ?? null])). nonce must be at least 16 bytes of hex; expiresAt, when present, must be a non-negative safe integer; and a lone UTF-16 surrogate anywhere in the invite's string fields is rejected (it would serialise to different bytes depending on the JSON encoder / platform, breaking the byte-exact digest).

Exact byte layouts (the ECDH construction + frozen vector, invite canonical bytes, kind/tag shapes, build/parse asymmetry) are in PROTOCOL.md.

kenspeckle makes narrow, honest claims — read SECURITY.md. The short version:

• Persona-scoping is a consumer obligation kenspeckle assists (discoverPresent enforces single-owner input) but cannot guarantee (no storage).
• Private annotations are never serialised (type-level + runtime); they ARE in the user's own encrypted backup (allowed — not a graph disclosure).
• kith/kin shared secrets are never published — toWire/serializeEntry strip sharedSecret structurally (0.2.0; previously only annotations was excluded). Cross-device sync of your OWN devices, which genuinely needs the secret to travel, uses the separately-named toSyncForm/serializeEntryForSync instead — never publish that output.
• NIP-05 is TOFU, not key-continuity — a key-change is propose-only, never auto-accept; require user confirmation. A rollback to a key already in previousPubkeys (a compromised/reverted domain re-serving an old key) is flagged distinctly and acceptKenRotation refuses it without an explicit { allowRevert: true }. acceptKenRotation also refuses an already-accepted rotation, so it can never double-append the current key into its own history.
• attributeSignature is replayable — use verifyKeyControl (fresh nonce) for impersonation resistance.
• Verify filter publications against a pinned key — check the Nostr sig AND the in-blob Schnorr sig; pass the raw wire event (a spread-mutated one carries a stale Symbol(verified) false-green). Since 0.2.0, parseFilterPublication/ parseFilterPublicationResult require opts: { namespace, serverId } — the in-blob signature is now bound to a context string (@forgesworn/tessera-kit 0.2.0) built only from opts, never from the event's own d-tag, which is what stops a relay/MITM from relabelling a substituted blob (see @forgesworn/tessera-kit PROTOCOL.md §4.3). The event's own d-tag is separately checked to equal that same context ('address-mismatch' on a mismatch). By default (requireAuthorIsSigner) the event's signer must also equal the in-blob signer, defence-in-depth alongside — not instead of — context binding.
• Discovery is opt-in + exit-able, but you cannot cryptographically verify removal from a pool you left. Real member privacy = per-context personas.
• tier:'kith' is a shape, not proof a bond happened. parseEntry/ importEntries accept any syntactically-valid 64-hex sharedSecret — kenspeckle cannot itself confirm a real deriveBondSecret ceremony occurred. Authenticate/encrypt whatever channel feeds parseEntry/importEntries; an unauthenticated sync source could otherwise escalate a one-way ken into a mutually-trusted-looking kith.

kenspeckle is the relationships brick of the Forgesworn / Signet ecosystem. It composes:

• @forgesworn/tessera-kit — the non-enumerable membership-presence filter kenspeckle/discovery is a thin layer over.
• spoken-token — the directional verification words the bond ceremony speaks.
• nostr-attestations — the co-signed bond assertions (kind-31000).
• nsec-tree — hierarchical persona key derivation (the ownerPubkey scoping; kenspeckle needs only the type relationship).

Consumed by signet-app (the user-facing identity app): it maps its Contact record onto KinEntry/KithEntry/ken, persists KindredEntry in IndexedDB, and keeps tier/badge enrichment app-side. The byte-exact ECDH (PROTOCOL.md §1) is what makes that migration lossless.

What has shipped, what is open and what each open item is blocked on: ROADMAP.md.

MIT.