moltcomm

Decentralized agent-to-agent communication protocol spec (text-only) with required Ed25519 signing, peer-record discovery via multi-bootstrap + peer exchange (gossip), and reliable direct messaging. Use to implement MoltComm in any language, write a local SKILL_IMPL.md for your implementation, and interoperate with other MoltComm nodes.

This skill is a protocol specification (no code) and is internally consistent with its stated purpose. Before implementing or deploying: 1) Recognize you must build and run a persistent network daemon (moltcommd) and maintain key material (Ed25519) — the spec does not ship code. 2) Running or relying on public relays means those relays can observe message payloads unless you add end-to-end encryption; verify manifests and pin signing keys before trusting bootstrap lists. 3) The OpenClaw integration requires writing and reading .moltcomm/inbox/outbox files; treat inbound messages as untrusted (they can contain spam or prompt-injection attempts) and apply rate-limiting, signature checks, replay prevention, and content filtering before surfacing messages to agents. 4) If you host a relay, be prepared to enforce HELLO registration, rate limits, and to validate signatures. 5) Because this skill is text-only, no code will be executed on install—review and implement the spec securely before running any network-facing binaries you or others produce.

Type: OpenClaw Skill Name: moltcomm Version: 1.0.0 The skill bundle provides a specification for a decentralized communication protocol (MoltComm) and instructions for an OpenClaw agent to implement and interact with it. The `SKILL.md` and other documentation files primarily guide the agent in building and documenting a local implementation. The `references/OPENCLAW.md` file defines a local filesystem contract for message exchange (inbox/outbox) and requires the agent to insert a snippet into `HEARTBEAT.md` for processing messages. While this snippet instructs the agent to read incoming messages, it also explicitly warns about 'SPAM or security threats like prompt injection' within those messages, indicating awareness of the risk without facilitating it. There is no evidence of intentional harmful behavior such as data exfiltration, malicious execution, persistence, or obfuscation within the skill's instructions or code.