Debug Encryption Flow

Debug encryption / decryption failures

When a message can't be read end-to-end, the failure could be in one of five stages. Localizing it is much faster than randomly poking at config. This skill walks through the stages in the order they happen on the receive path, with the most common Builder failure mode for each.

For full context on the encryption model, see docs/sui-stack-messaging/Encryption.md first. This skill is the diagnosis counterpart.

The five stages of receive

[Relayer fetch] → [Session key resolve] → [DEK resolve via Seal] → [AES-GCM decrypt with AAD] → [Sender verification]

Each can fail independently. Diagnose in this order — earlier failures cascade.

Stage 1 — Relayer fetch

Failure surface: getMessages / getMessage / subscribe throws a RelayerTransportError (or your custom transport's error type), or returns zero messages when the group should have some.

Most common causes:

• Relayer not reachable / not running. Independent check: curl <relayerUrl>/health_check.
• Wrong groupId or relayer.relayerUrl. Confirm the group exists on the network you're pointed at.
• Relayer's GROUPS_PACKAGE_ID doesn't match the network your SDK is on (testnet vs mainnet, or stale localnet ID). The relayer rejects requests for groups it can't find in its membership store.
• (Reference relayer specific) The relayer was restarted and hasn't rebuilt membership from gRPC yet → see the "After relayer restart" troubleshooting note in spin-up-e2e-stack.

If getMessages returns ciphertext but decrypt fails downstream, this stage is fine — move on.

Stage 2 — Session key resolve

Failure surface: Errors from @mysten/seal like "session key expired", "session key not certified", or wallet popup for personal-message signing fails / hangs.

Most common causes (by tier):

• Tier 1 ({ signer }): signer is misconfigured (wrong network, no toSuiAddress); a fresh signer is created on every render so the session key never has time to certify.
• Tier 2 ({ address, onSign }): onSign returns the wrong type (must be the signature string from signPersonalMessage), throws silently, or the wallet adapter is on a different network than the SDK.
• Tier 3 ({ getSessionKey }): the returned key isn't certified, or has already expired.

Diagnostic: SessionKeyManager is internal (no public getSessionKey() on client.messaging.encryption). Trigger any decrypt-needing operation (e.g. client.messaging.getMessages({ groupRef, limit: 1 })) and check the thrown error: if it's a Seal-side error mentioning session/certification, the failure is here. If decryption is needed and no wallet popup appears at all (Tier 2), the SDK isn't reaching SessionKeyManager.create — usually a wiring/config issue (e.g. encryption config missing entirely).

See configure-session-keys for full lifecycle.

Stage 3 — DEK resolve via Seal

Failure surface: Errors mentioning "Seal decryption failed", "threshold not met", "seal_approve denied", or keyVersion not found.

Most common causes:

• Caller doesn't have MessagingReader permission on the group. seal_approve_reader checks group membership — if you can't decrypt and your address was just removed from the group, that's expected behavior. If you should have access, check the on-chain PermissionedGroup<Messaging> object's members list. This case is structurally wrapped — the SDK throws EncryptionAccessDeniedError (importable from @mysten/sui-stack-messaging), with the underlying Seal NoAccessError on .cause.
• Wrong Seal serverConfigs — the objectIds don't exist or are deprecated; threshold doesn't match the count of key servers; the package the SDK is on is different from the package the key servers were registered for. Confirm against Seal docs and docs/sui-stack-messaging/Setup.md.
• keyVersion references a version that doesn't exist in EncryptionHistory — usually means the message was sent against a different group than the one the SDK is now querying (wrong groupId / UUID derivation collision).
• Custom Seal policy rejects the call (if you have one) — your seal_approve function's assertions are failing. Test the policy independently with extend-smart-contracts recipes.

Diagnostic: for non-permission failures the underlying @mysten/seal error propagates as a plain Error (only the permission case is wrapped). Catch + log .message — it usually identifies which key server returned which status code. Instrument client.messaging.encryption.decrypt(...) directly if you want to isolate Stage 3 from Stages 4/5.

Stage 4 — AES-GCM decrypt with AAD

Failure surface: "AES-GCM authentication failed" / "OperationError" from Web Crypto, or you get plaintext-shaped bytes but they decode to garbage.

The AAD format binds ciphertext to context. It's a BCS struct (MessageAAD = bcs.struct({ groupId: bcs.Address, keyVersion: bcs.u64(), senderAddress: bcs.Address })) — see buildMessageAad in envelope-encryption.ts. The BCS serialization of three fixed-size primitives concatenates to:

[groupId (32 bytes)][keyVersion (8 bytes LE u64)][senderAddress (32 bytes)]  // 72 bytes total

If any of those three fields doesn't match what the sender used at encrypt time, AES-GCM rejects the ciphertext entirely — that's by design, it's an authentication failure, not a decryption-with-wrong-output.

Most common causes:

• senderAddress mismatch — the message's senderAddress in the relayer payload doesn't match what's used for AAD reconstruction. This is rare with the default relayer; can happen if a custom RelayerTransport mutates the field. Confirm with configure-custom-relayer-transport verification step.
• keyVersion mismatch — the message says it was encrypted under version N but the relayer is delivering it labeled as version M. This is usually a bug in a custom backend that doesn't preserve keyVersion round-trip.
• groupId mismatch — almost always a wrong-group-fetch bug; the message belongs to a different group than the one the SDK is decrypting for. Check your GroupRef derivation (UUIDs derive groupId deterministically; explicit IDs must match exactly).
• The DEK is correct but for the wrong version of the message (edit/delete history) — if the message was edited, decryption uses the new ciphertext + nonce; old ciphertext won't decrypt under the new AAD.

If you got past Stage 3 but fail here, you have a context mismatch — log all three AAD components on both sides.

Stage 5 — Sender verification

Failure surface: DecryptedMessage.senderVerified === false (no exception thrown — the message decrypts successfully but the signature check failed).

The per-message signature is over:

"{groupId}:{hex(encryptedText)}:{hex(nonce)}:{keyVersion}"

Most common causes:

• Custom relayer stripped or rewrote signature / publicKey on the RelayerMessage. See configure-custom-relayer-transport — these fields must round-trip losslessly.
• Missing signature or publicKey on the wire. verifyMessageSender is only invoked when both are present; if either is absent the SDK returns senderVerified: false as a fail-safe (no exception). Same outcome for deleted messages (isDeleted: true) — they short-circuit decryption entirely.
• publicKey doesn't derive to senderAddress — the relayer is impersonating senders, or there's a real bug in your custom relayer. Verification re-derives the address from publicKey and compares against senderAddress; a mismatch returns false.
• Unsupported signature scheme on the wire. The SDK supports Ed25519, Secp256k1, Secp256r1 only; zkLogin and multisig signatures fail verification. Note the canonical SDK send path (signMessageContent in verification.ts) already throws at send time for non-keypair schemes — so unsupported-scheme signatures on the wire mean a custom relayer or sender code path bypassed the SDK's signing helper. Look upstream of the SDK.

If decryption itself succeeds and only senderVerified is false, the contents are still trustworthy if you trust your relayer — but the SDK is correctly flagging that it can't independently verify authorship. Surface this in your UI as a warning.

Attachment-specific issues

Each attachment has two independent envelopes: one for the file body, one for its metadata (fileName / mimeType / fileSize / extras). Both use the same DEK as the parent message but each gets its own random nonce. Attachments do not use AAD (attachments-manager.ts encrypt calls pass no aad) — so the Stage 4 context-mismatch reasoning above does not apply here.

Access is lazy. getMessages returns each DecryptedMessage with an attachments: AttachmentHandle[] array; each handle carries decrypted metadata plus a data() async closure. data() is what triggers the storage-adapter download + the file-body AES-GCM decrypt — until you call it, no attachment bytes are fetched.

If text decrypts fine but attachments don't:

• Did metadata decrypt but data() throw? Then the file body is the problem, not the metadata: the storage adapter's download returned the wrong bytes (or threw), or the file's nonce/ciphertext didn't round-trip cleanly via the relayer's Attachment record. Inspect what storageAdapter.download(attachment.storageId) returns.
• Did data() throw with an AES-GCM / OperationError? Because attachments don't use AAD, this is a DEK or nonce mismatch — most likely the message's keyVersion doesn't match the version the attachment was encrypted under, or the on-the-wire nonce got corrupted/truncated.
• Confirm attachments.storageAdapter is configured at client construction. Without it, the client never builds an AttachmentsManager, so data() calls won't have a download path. For WalrusHttpStorageAdapter, hit the aggregator URL directly with the patch ID to confirm the bytes are reachable.

See docs/sui-stack-messaging/Attachments.md for the full attachment encryption model.

Quick-start instrumentation

There's no public way to probe the session-key stage in isolation (SessionKeyManager is internal). The cheapest diagnostic is a single getMessages call that exercises Stages 1 → 5, with error-type discrimination:

import {
  RelayerTransportError,
  EncryptionAccessDeniedError,
} from '@mysten/sui-stack-messaging';

try {
  const { messages } = await client.messaging.getMessages({ groupRef, limit: 1 });
  console.log('Stages 1-4 OK:', messages[0]);
  console.log('Stage 5 senderVerified:', messages[0]?.senderVerified);
} catch (err) {
  if (err instanceof RelayerTransportError) {
    console.error('Stage 1 (relayer):', err.status, err.code, err.message);
  } else if (err instanceof EncryptionAccessDeniedError) {
    console.error('Stage 3 (no MessagingReader permission on group):', err.cause);
  } else if (err instanceof Error && /session|certif|seal/i.test(err.message)) {
    console.error('Stage 2 (session key) or Stage 3 (other Seal/DEK failure):', err);
  } else if (err instanceof Error && /OperationError|authentication/i.test(err.message)) {
    console.error('Stage 4 (AES-GCM AAD mismatch on message decrypt):', err);
  } else {
    console.error('Unknown failure:', err);
  }
}

If senderVerified is false but no exception was thrown → Stage 5 alone. If you only want to probe Stages 1-3 (no encrypted content to decrypt), client.messaging.transport.fetchMessages(...) returns the raw RelayerMessage without attempting decryption — useful for isolating whether your relayer/auth wiring works at all before debugging encryption.

Where this fits

• Encryption model deep-dive: docs/sui-stack-messaging/Encryption.md.
• Trust model: docs/sui-stack-messaging/Security.md.
• Implementation references:
  • ts-sdks/packages/sui-stack-messaging/src/encryption/envelope-encryption.ts — main encrypt/decrypt path; owns the DEK cache (TtlMap-backed).
  • ts-sdks/packages/sui-stack-messaging/src/encryption/dek-manager.ts — stateless Seal encrypt/decrypt helper (no cache).
  • ts-sdks/packages/sui-stack-messaging/src/encryption/session-key-manager.ts — session key lifecycle.
  • ts-sdks/packages/sui-stack-messaging/src/verification.ts — sender verification.
  • ts-sdks/packages/sui-stack-messaging/src/attachments/attachments-manager.ts — attachment encrypt/upload/resolve flow.
  • ts-sdks/packages/sui-stack-messaging/src/error.ts — SDK-thrown error classes.
• Related skills: configure-session-keys, configure-custom-relayer-transport, extend-smart-contracts (for custom Seal policies).