RFD: Streamable HTTP & WebSocket Transport

[federicociner]

Draft implementation of the Streamable HTTP & WebSocket Transport RFD.

Reference implementation aligns with the RFD and most of the proposed changes in the Rust SDK.

[benbrandt]

@federicociner I'll need to dig in for sure, it's a big one! But I wonder if there is a way we can make it clearer this is experimental/unstable while we figure out how this will all work?

Open to multiple solutions (either in the filenames/exports, a separate package, something else)

[federicociner]

@federicociner I'll need to dig in for sure, it's a big one! But I wonder if there is a way we can make it clearer this is experimental/unstable while we figure out how this will all work?

Open to multiple solutions (either in the filenames/exports, a separate package, something else)

@benbrandt yes agreed. I think the main reasonable options are:

1. Experimental sub-path exports e.g.

@agentclientprotocol/sdk/experimental/http-client
@agentclientprotocol/sdk/experimental/ws-client
@agentclientprotocol/sdk/experimental/server
@agentclientprotocol/sdk/experimental/node

and we can just update the exports to

{
  "exports": {
    ".": {
      "types": "./dist/acp.d.ts",
      "import": "./dist/acp.js",
      "default": "./dist/acp.js"
    },
    "./experimental/http-client": {
      "types": "./dist/http-stream.d.ts",
      "import": "./dist/http-stream.js",
      "default": "./dist/http-stream.js"
    },
    ...
  }
}

2. Unstable prefixes for all symbols e.g.

unstable_createHttpStream
unstable_createWebSocketStream
UnstableAcpServer
unstable_createNodeHttpHandler
unstable_createNodeWebSocketUpgradeHandler
...

My personal preference is Option 1, since there's less churn and breaking renames later on when we stabilise the spec.

WDYT?