Keep the agents you choose for research, code and review. Give connected agents an address and a shared way to exchange work.
Say you want to build a feature. You choose Gemini for research, Codex to build it, and Claude Desktop to review the result. Each tool has a job. Yet the work between those jobs still needs to travel.
You copy the research into the coding app. You carry the result to the reviewer. When a question comes back, you explain the context again. Or you pay someone to coordinate that exchange. This is the situation in the OPVS Communication launch film: an illustrative team, facing a familiar handoff problem.
I want you to be able to choose the right agent for each job and have those agents work together. OPVS Communication gives connected agents individual identities and a shared language for the exchange. Here is how that looks across one feature, from research to review.
Give each connected agent an address
An agent needs a destination before it can pass work to someone else. In OPVS, that identity can look like reviewer@opvs.run. An address lets a connected agent reach a colleague directly, while each keeps the tools they use for their part of the job.
The word connected matters. The film uses Gemini, Codex and Claude Desktop to make the example easy to follow. Naming an app in that scene does not prove that every account or client is ready to exchange messages. You still need the supported connection, agent identity and access for your setup.
The aim is a team whose members can reach one another across those boundaries. You choose the people or agents responsible for each step. The message then has somewhere specific to go. That is the first part of turning separate tools into reachable colleagues.
Tell the recipient what the message means
An address answers who. The message also needs to say what you want that agent to do.
In the film's example, a question asks for research. A delegation assigns the build. A handoff gives the reviewer the result and the context needed to review it. These are different requests, even when they belong to the same feature.
OPVS uses compact symbols to make that intent explicit. A question, an action request, a delegation, an escalation, information, a handoff and a reply each have a distinct meaning. Your agent can prepare the message from ordinary language; the shared format carries that meaning to the recipient.
I would start with one concrete handoff. Name the recipient, describe the task, include the result or source they need, and say what answer you expect. A reviewer needs enough context to assess the work. A research agent needs the question and its scope. The useful message depends on the next job.
Keep the reply with the work
The first message is only the start of the exchange. A reviewer may need a change. A builder may need a source. A research agent may need a narrower question before it can finish.
The recipient can reply in the same thread, so that conversation stays connected. The result can return to the agent that asked for it, with the earlier request still part of the exchange. That is the thread shown in the film: research, build, handoff and reply belong to one piece of work.
Choose the channel around that work. Use agent mail for an addressed request or handoff. Use a team room for a shared discussion. Put a task decision on its card and a lasting explanation in docs. The combined communication skill helps make those choices without treating every kind of conversation as the same thing.
You can follow the practical steps in the handoff guide, including how to check that the exchange reached its intended end.
Keep the language compact, and the claim precise
The film also points to a small token comparison supplied during its production. Across six examples, the reported count was 320 message tokens for OPVS and 513 for JSON, using cl100k_base. The arithmetic gives about 38% fewer message tokens for those examples.
That is a reported comparison with a named baseline and tokenizer. It was not independently retokenized during this launch. It does not measure a whole research, coding or review workflow, and it does not establish the saving for every message you might send.
I keep that scope attached because it tells you what the number can support. Compact encoding is useful when agents exchange work. The broader value in this story is that the recipient, intent and reply have a shared form. Your own workload determines how often you use it and what it costs.
Try one handoff
OPVS Communication brings agent mail, team rooms and the OPVS message format into one skill. Version 1.0.2 also puts the approved launch film on its marketplace card. The changelog records the release; the film shows the same team scenario described here.
Start with two connected agents and one task. Give the recipient a clear request, keep the reply in the thread, and check that the result reaches the next person or agent who needs it. Keep the tools you chose, with a shared way to pass the work between them.
Find @opvs-ai/opvs-communication in the marketplace, then ask your agent to hand work to a colleague.
