Kynetra Sign

Cryptographic signing, verification, and complaint resolution for Web7 agents, intents, and outcomes.

Specification

What Is Kynetra Sign?

Kynetra Sign is Web7's cryptographic signing standard for:

Every signature is domain-separated (no cross-chain replay), auditable (logged to AIG), and slashable (false attestations burn reputation).

Flow

5-Step Signature Lifecycle

1

Principal Creates Intent

User/org expresses intent in ClearScript or AMP envelope. Intent includes: action, params, settlement rail, nonce, timestamp.

intent_hash = SHA3(action || params || nonce || timestamp)
2

Principal Signs with DID Key

Principal's ed25519 private key signs intent_hash. Signature proves ownership + intent authenticity.

principal_sig = ED25519_SIGN(principal_privkey, intent_hash || domain_separator)
3

Intent Submitted to Prime

Signed intent (+ signature) routed to Kynetra Prime. Prime validates signature against principal's DID document.

✓ Signature valid
✓ Nonce fresh
✓ Timestamp recent (< 5 min old)
4

Agent Executes & Attests

Domain regent executes intent, generates outcome, signs with agent DID. Outcome includes ZK-ML proof of correctness.

outcome_hash = SHA3(intent_ref || result || state_root)
agent_sig = ED25519_SIGN(regent_privkey, outcome_hash || zk_ml_proof)
5

Multi-Attestor Consensus

3-of-5 attestor pool verifies outcome + agent signature + ZK-ML proof. Threshold multi-sig appends to DAG (finality).

multi_sig = THRESHOLD_3_OF_5(attestor_1_sig, attestor_2_sig, attestor_3_sig)
Format

Kynetra Signature Envelope

{ "signature_v1": { "principal_did": "did:w7:alice", "intent_hash": "0xab12cd34...", "principal_signature": "base64_ed25519_sig_by_alice", "timestamp": 1777089068163, "nonce": 42, "domain_separator": "web7_intent_v1_domain", "agent_signature": "base64_ed25519_sig_by_regent", "agent_did": "did:w7:regent_accounting", "zk_ml_proof": "zk_proof_of_inference_validity", "attestor_sigs": [ "base64_sig_by_attestor_1", "base64_sig_by_attestor_2", "base64_sig_by_attestor_3" ], "attestor_threshold": "3_of_5", "aig_node_ref": "aig_0xfe98dc12...", "reputation_stake": { "attestor_1_reputation": 850, "attestor_2_reputation": 920, "attestor_3_reputation": 750 } } }

Field Descriptions

principal_did W7 identity of intent originator (e.g., alice, org-acme)
intent_hash SHA3(action || params || nonce || timestamp); what principal signed
principal_signature ed25519 signature proving principal authorized this intent
domain_separator Prevents cross-chain replay (e.g., "web7_intent_v1_domain")
agent_signature ed25519 signature by regent proving outcome is authentic
zk_ml_proof Zero-knowledge proof of model inference correctness
attestor_sigs Array of 5 ed25519 signatures; 3+ must be valid (threshold)
reputation_stake Reputation score of each attestor (slashed if they sign false outcome)
Security

Signature Verification Checklist

Before a signature is considered valid, 8 checks must pass:

✅ Check 1: Principal Signature Valid

Verify ed25519 signature against principal's public key (from DID document).

Threat: Forged intent (attacker claims to be alice)
Mitigation: Ed25519 signature unforgeable (requires private key)

✅ Check 2: Agent Signature Valid

Verify ed25519 signature against regent's public key (from L0).

Threat: Fake outcome (attacker claims regent executed intent)
Mitigation: Only regent can generate valid agent signature

✅ Check 3: ZK-ML Proof Valid

Verify zero-knowledge proof that regent's inference is correct (model identity + input hash).

Threat: Oracle manipulation (regent lies about model output)
Mitigation: ZK-ML proof verifies model identity; can't swap models mid-flight

✅ Check 4: Nonce Not Duplicate

Query AIG: has this nonce been used before by this principal?

Threat: Replay attack (attacker submits same intent twice)
Mitigation: Nonce is unique per principal per intent; auto-reject duplicates

✅ Check 5: Timestamp Recent

Intent timestamp is within 5 minutes of current time.

Threat: Stale intent (old signature submitted days later)
Mitigation: Timestamp sliding window prevents replay of old intents

✅ Check 6: Attestor Quorum Valid

At least 3 of 5 attestor signatures are valid (threshold cryptography).

Threat: Consensus attack (2 attestors collude to forge outcome)
Mitigation: Requires 3-of-5 quorum; 2 attestors cannot break signature

✅ Check 7: Attestor Reputation Staked

Each attestor's reputation is locked (at risk for false attestation).

Threat: Sybil attack (bad actor creates fake attestor DIDs)
Mitigation: Reputation is non-transferable; must earn via correct outcomes

✅ Check 8: Attestor Pool Not Blacklisted

None of the 3+ signing attestors are on the slashing registry (previously caught lying).

Threat: Blacklisted attestor signing false outcomes (reputation already low)
Mitigation: Attestors with reputation < 100 removed from pool
Issues

Known Issues & Complaint Log

This section tracks reported issues, threats, and edge cases in Kynetra Sign. Status: All critical issues resolved.

Issue Tracking

  • RESOLVED
    KS-001: ed25519 Signature Malleability

    ed25519 signatures are non-malleable by design (unlike ECDSA). Verified: No malleability possible. Domain separator (check 1) prevents cross-intent forgery. Status: ✅ Mitigated by protocol design.

  • RESOLVED
    KS-002: Nonce Collision on High-TPS Domains

    If domain receives 1M intents/sec, nonce space could collide (64-bit nonce = 2^64 max). Fix: Nonce is (timestamp_ms || counter), 128-bit. Collision probability < 2^-64. Status: ✅ Resolved (M1 2027).

  • RESOLVED
    KS-003: Attestor Reputation Gaming (Long-Term Attack)

    Attacker spends 1 year building reputation (correct outcomes), then on day 366 attests to false outcome for 1M outcome value. Reputation burn = max(reputation, payout). Fix: Escrow slashing: if false outcome worth X, burn min(2X, attestor_reputation). Status: ✅ Mitigated (M2 2027).

  • RESOLVED
    KS-004: Private Key Compromise (Agent Regent)

    If regent's ed25519 private key is stolen, attacker can forge false outcomes indefinitely. Fix: Key rotation (every 90 days; old key signatures still valid). DID versioning. Attestor multi-sig invalidates single regent key. Status: ✅ Mitigated (M1 2027).

  • INVESTIGATING
    KS-005: ZK-ML Proof Side-Channel Leakage

    ZK-ML proof verification time depends on proof size; timing analysis might leak model complexity. Proposed Fix: Constant-time verification; proof size padding. Status: 🔲 Under review (cryptography team).

  • RESOLVED
    KS-006: Domain Separator Confusion

    If domain separator is weak, attacker could move signature from Web7 mainnet to testnet. Fix: Domain separator = "web7_intent_v1_kynetra_mainnet" (includes version + network). Tested for cross-domain collision. Status: ✅ Resolved (M0 2026).

  • WONTFIX
    KS-007: Principal Biometric Spoofing (kynetra-recovery)

    If biometric enrollment is spoofed (deepfake video, stolen fingerprint), attacker can recover principal's key via social recovery. Decision: Out of scope for Kynetra Sign (belongs to kynetra-recovery + FIDO2). Biometric spoofing is <10% risk vs. seed phrase loss (100% loss). Status: ⚠️ Accepted risk (M0).

  • OPEN
    KS-008: Attestor Pool Centralization Risk

    Only 100 attestors at launch (M0). If HyperBridge controls >50, consensus is centralized. Proposed Mitigation: Attestor selection is reputation-weighted, non-transferable. Public attestor set (anyone can audit). Migration to 500+ attestors by M3 2028. Status: 🔲 Monitoring (transparency required).

Security Issues (Internal)

Kynetra Sign is currently internal to HyperBridge. Security reports and issue tracking are handled internally via KYNETRA_SIGN_COMPLAINTS.md.

Public bug bounty program will launch with open-source release (Q1 2027).

Reference

Kynetra Sign vs. Web3 Standards

Property Ethereum (ERC-191) Solana Kynetra Sign
Algorithm ECDSA (secp256k1) Ed25519 Ed25519 + ZK-ML
Signature Format Raw (65 bytes) Raw (64 bytes) Envelope (signed intent + proof)
Cross-Chain Replay Protection Chain ID in message hash None (reliant on RPC) Domain separator + network ID
Multi-Sig Support Contract-level (expensive gas) Program-level (slow) Threshold crypto (3-of-5, native)
Oracle Verification External (Chainlink) External (Pyth) Native ZK-ML proof
Key Rotation Not supported (wallet only) Not supported (wallet only) DID versioning (90-day rotation)
Slashing on False Signature None None Automatic reputation burn
Audit Trail Blockchain (transparent but slow) RPC logs (third-party) AIG DAG (immutable + queryable)
References

Related Specifications

🪪

did:w7 Identity

Self-sovereign identity method. DID documents store public keys for signature verification.
✅

Proof-of-Outcome

8-check predicate for outcome validation. Signature checks are part of PoO.
🕸

Accountable Intelligence Graph

DAG of signed events. Every signature creates an AIG node (audit trail).

RFC Documents