Skip to content

cross chain spec

github-actions[bot] edited this page Sep 16, 2026 · 4 revisions

title: Cross-Chain Reputation Sync [PLANNED] created: 2026-07-09 updated: 2026-09-15 type: concept tags: [layer-2] confidence: low source_files:

  • contracts/src/oracle/CCIPReputationBridge.sol
  • README.md

[PLANNED] — not built, not deployed, not wired in. The root README.md's "Decentralization path" lists cross-chain reputation (syncing AIS across Base/Arbitrum/Ethereum) as a real future step, not a current capability. This page replaces the old wiki's cross-chain-spec.md, which described a "Canonical Reputation Registry" + "Satellite Synchronizers" design assuming the pre-rewrite singleton ReputationRegistry model — that model is gone (see agent primitives), so the old design doesn't apply as written.

Table of contents

Current state: an honest, documented gap

contracts/src/oracle/CCIPReputationBridge.sol exists in the codebase and its per-agent clone resolution has been reworked: both send and receive resolve the agent's ReputationRegistry through XibalbaAgentRegistry, and an unregistered agent fails closed. The bridge remains explicitly unwired for production: it is not deployed by Deploy.s.sol, and each agent must opt-in by granting the bridge BRIDGE_ROLE on its own clone while operators configure trusted peer bridges and a real CCIP lane. The contract and its tests are therefore implemented but still [PLANNED] as a deployed cross-chain capability.

What a real design would need to solve

Not built, but the shape a working version would have to address: per-agent clone resolution (above) instead of a single hub registry; a bridge-agnostic messaging layer (CCIP, the contract's namesake, or an alternative) rather than vendor lock-in; a minimum confirmation delay to mitigate bridge-reorg risk before a synced score is trusted on the destination chain. None of this has an implementation to point at.

Related: AIS, agent primitives, contracts.

Clone this wiki locally