Agent Session Library
Node.js client for the agentd socket, plus a small demo window and a simulated test-driven-development runner.
robos-agent-session is mainly a library. Its window exists to demonstrate the API.
The library
index.js exports RobOSAgentSession, AutonomousEDDRunner and EDD_PHASES. RobOSAgentSession talks to Agent Daemon over the Unix socket ~/.config/robos/agentd/agentd.sock (override with the socketPath option). If the socket is missing, calls resolve to { ok: false, error: "Socket not found at ..." }. You can also pass an in-process daemon object to skip the socket.
| Method | What it sends |
|---|---|
spawnAgentSession(taskId, options) | spawn |
listAgentSessions() | list; returns an array |
inspectAgentSession(taskId) | inspect |
sendAgentCommand(taskId, command) | appendLog with the line [COMMAND_DISPATCH] <command>. It records the command in the session log; it does not execute it. |
terminateAgentSession(taskId) | terminate |
wipeAllSessions() | wipeAll |
runEDD(config) | Runs the local EDD runner (below). |
The object is an EventEmitter and emits session:spawned, session:command, session:terminated and session:wiped.
The EDD runner
lib/edd-runner.js steps through the phases ingestion, red verification, implementation, green verification, regression check and completed, and logs each one. The red and green checks are simulated: the runner returns fixed results and only calls fabric.dispatchRequest if you pass a fabric object. Treat it as a scaffold for the phase model, not a test runner. Knowledge Graph Explorer loads it.
The demo window
main.js opens RobOS Agent Session Client Demo. It starts its own in-process agent daemon, so it does not need robos-agentd to be running. The window has + Spawn via Library, Refresh and Send Command controls, a session list and an event log. There is no .desktop launcher.
Tests
npm test in the package runs node --test.