Node-based terminal manager for AI coding agents — tmux-backed terminals and parallel agent sessions as draggable nodes on an infinite pan/zoom canvas. macOS, Linux, and a browser Server Edition.
TypeScript adhdagent-orchestrationai-agentscanvasclaude-codecodexcoding-agentsdeveloper-toolselectrongeminilinuxmacosmultiplexeropencodeparallel-agentsterminalterminal-emulatorterminal-multiplexertmuxworkspace-manager
// scripts/chat-sdk-spike.mjs
// Spike: verify @anthropic-ai/claude-agent-sdk works when spawned from the Electron
// binary in node mode (same runtime as our main process). Run:
// ELECTRON_RUN_AS_NODE=1 ./node_modules/.bin/electron scripts/chat-sdk-spike.mjs
// Success = token deltas print live, then a result line with cost/usage.
//
// ── SPIKE RESULT (Task 1 — Task 5 copies this verbatim) ──────────────────────
// PASS with NO env/executable overrides needed.
// query() streamed from the Electron-as-node runtime out of the box:
// [init] with session id + 127 slash commands, live text_delta tokens, then a
// [result] success line with cost/usage.
// Notes for Task 5 (ChatDriver in main):
// - ELECTRON_RUN_AS_NODE=1 was present in the PARENT env and inherited by the
// SDK's spawned CLI subprocess; the SDK still worked correctly. No need to
// strip it from the child env, and no `env` sanitization was required.
// - No `pathToClaudeCodeExecutable` override needed — the SDK found/spawned its
// bundled CLI on its own. (If a future runtime can't resolve it, the fallback
// is `options.pathToClaudeCodeExecutable = <result of \`which claude\`>`.)
// - Runs under `app.isPackaged === false` dev too; the production main process
// is the same Electron-node runtime, so the same call site applies.
// ─────────────────────────────────────────────────────────────────────────────
import { query } from '@anthropic-ai/claude-agent-sdk'
async function* prompts() {
yield {
type: 'user',
message: { role: 'user', content: 'Reply with a two-sentence joke about terminals.' }
}
// Keep the iterable open briefly so the turn can finish, then end it.
await new Promise((r) => setTimeout(r, 60_000))
}
const q = query({
prompt: prompts(),
options: {
cwd: process.cwd(),
includePartialMessages: true,
permissionMode: 'default',
allowedTools: [] // pure-text turn; no permissions should fire
}
})
for await (const msg of q) {
if (msg.type === 'stream_event') {
const delta = msg.event?.delta
if (delta?.type === 'text_delta') process.stdout.write(delta.text)
} else if (msg.type === 'system' && msg.subtype === 'init') {
console.log('[init] session', msg.session_id, 'slash:', (msg.slash_commands ?? []).length)
} else if (msg.type === 'result') {
console.log('\n[result]', msg.subtype, 'cost=', msg.total_cost_usd, 'usage=', msg.usage)
break
}
}
process.exit(0)