Skip to content

an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions

revert(workers): drop the original #159 spawn/teardown commit

Chaitanya Giri73e4fefcbf0c
diff --git a/src/main/index.ts b/src/main/index.ts
index d80f18c0..1690c290 100644
--- a/src/main/index.ts
+++ b/src/main/index.ts
@@ -413,21 +413,11 @@ const preservedWorktrees = new Map<string, PreservedWorktree>();
* per dead PTY.
*
* Idempotent: guarded on map presence and the already-idempotent
- * `hive.setArchived`, so a double call is a harmless no-op. NOTE: an explicit
- * `ptyManager.kill()` does NOT reach here via onExit — kill() deletes the
- * session synchronously, so node-pty's later async exit callback fails the
- * session-identity guard and is swallowed. Every kill site must therefore call
- * teardownPty itself right after the kill (all of them do). Best-effort — every
- * step is wrapped so a teardown error can never crash the caller (an IPC
- * handler or node-pty's onExit).
+ * `hive.setArchived`, so the second call (kill() also makes node-pty fire
+ * onExit) is a harmless no-op. Best-effort — every step is wrapped so a teardown
+ * error can never crash the caller (an IPC handler or node-pty's onExit).
*/
function teardownPty(id: string): void {
- // Main cards an ephemeral worker on the floor when it spawns, so main has to
- // un-card it too — otherwise a released worker leaves a ghost the renderer's
- // nudge/drain loops keep polling. Read the flag before the branches below
- // clear it. Renderer-owned agents are deliberately untouched: the renderer
- // owns those cards and keeps a dead one so the user can Restart & Continue.
- const wasWorker = liveWorkers.has(id);
// 0) Revoke this id's broker capability (if any). Idempotent + harmless for a
// non-worker PTY; ensures a dead worker's token can never reach an integration.
try { integrationBroker.revoke(id); } catch { /* best-effort */ }
@@ -467,9 +457,6 @@ function teardownPty(id: string): void {
// A worker whose isolation failed (non-repo cwd) has no worktree to gate above —
// still clear its tracking entry so the controller stops watching a dead PTY.
if (liveWorkers.has(id)) liveWorkers.delete(id);
- if (wasWorker) {
- try { liveWebContents()?.send('hive:agentArchived', { id }); } catch { /* window gone */ }
- }
syncKeepAwake();
}
@@ -4404,29 +4391,8 @@ async function processSpawnRequest(filePath: string): Promise<void> {
const cwd = typeof raw.cwd === 'string' && raw.cwd.trim() ? expandTilde(raw.cwd) : '';
if (!cwd || !existsSync(cwd)) { fail(`"cwd" missing or not found (${cwd || 'unset'})`); return; }
- const cfgSpawn = readConfig();
- let command = typeof raw.command === 'string' && raw.command.trim() ? raw.command.trim() : (cfgSpawn.defaultCommand ?? 'claude');
- // Inherit the app's auto (skip-permissions) mode when the request takes no
- // stance of its own: a headless worker has no human to click through tool
- // prompts, so without the flag it stalls at the first ask until the idle
- // reaper kills it. An explicit --permission-mode in the request always wins,
- // and non-claude commands are left untouched.
- if (cfgSpawn.autoMode && inferAgentProvider(command, raw.provider) === 'claude' && !command.includes('--permission-mode')) {
- command += ' --permission-mode bypassPermissions';
- }
- // god authors `command` as a full command LINE ("claude --model … --permission-mode …"),
- // but the PTY layer takes ONE executable name (resolveCommand) plus argv — the
- // unsplit line made node-pty exec a binary literally named like the whole
- // string → ENOENT → the worker died within ~1s of spawning while its request
- // archived as .done (this killed both flag-carrying Ryan spawns on 2026-08-16;
- // only the bare-`claude` one lived). Split it here — same quoting rules as the
- // renderer's tokenizeCommand — and hand the flags over as argv.
- const tokens: string[] = [];
- const tokenRe = /"([^"]*)"|'([^']*)'|(\S+)/g;
- let tok: RegExpExecArray | null;
- while ((tok = tokenRe.exec(command)) !== null) tokens.push(tok[1] ?? tok[2] ?? tok[3]);
- const bin = tokens[0] || command;
- const cmdFlags = tokens.slice(1);
+ const command = typeof raw.command === 'string' && raw.command.trim() ? raw.command.trim() : (readConfig().defaultCommand ?? 'claude');
+ const bin = command.split(/\s+/)[0] || command;
// Missing-CLI → FAIL FAST. A headless worker has no human to watch an installer,
// so we never run the cc49e1e install banner here — we reject and tell god.
if (!ptyManager.isCommandAvailable(bin)) { fail(`engine CLI "${bin}" is not installed`); return; }
@@ -4455,44 +4421,21 @@ async function processSpawnRequest(filePath: string): Promise<void> {
brokerEnv.MD_BROKER_TOKEN = token;
}
const spawnOpts: AgentSpawnOptions = {
- id: workerId, cwd, command: bin, cols: 120, rows: 32,
- // A separate `model` field only applies when the command line didn't pick a
- // model itself (spawnAgentCore likewise skips its default-model injection
- // when argv already carries --model).
- args: [...cmdFlags, ...(raw.model && !cmdFlags.includes('--model') ? ['--model', raw.model] : [])],
+ id: workerId, cwd, command, cols: 120, rows: 32,
+ args: raw.model ? ['--model', raw.model] : [],
hive: meta, isolate, provider: raw.provider, env: brokerEnv
};
- let res: { ok: boolean; error?: string; worktreePath?: string };
+ let res: { ok: boolean; error?: string };
try {
+ // Output routes to the primary window (no renderer evt here). Workers are
+ // headless-by-design — they reply to Slack + report to god, not a watching human.
res = await spawnAgentCore(spawnOpts, liveWebContents());
} catch (e) {
res = { ok: false, error: String(e) };
}
if (!res.ok) { integrationBroker.revoke(workerId); fail(`spawn failed — ${res.error ?? 'unknown error'}`); return; }
- // Card the worker on the floor. The renderer's roster is renderer-owned and is
- // only mutated by renderer-initiated hires, so without this broadcast a worker
- // spawned from a spawn-request is invisible in the GUI — and, far worse, it is
- // invisible to the renderer loops that actually DRIVE an agent. The objective
- // below is dispatched through the inbox, and the loop that turns inbox mail
- // into a terminal nudge (useHive #3) plus the queue drain that types it in
- // (#4) both iterate the renderer's agent list. An uncarded worker therefore
- // sat at its prompt with its objective unread — the whole feature was inert.
- // Same descriptor as the voice-hire path; the renderer's handler is idempotent
- // and bails on a duplicate id, so this can never double-card an agent.
- try {
- liveWebContents()?.send('hive:agentSpawned', {
- id: workerId,
- name: meta.name,
- provider: spawnOpts.provider ?? meta.provider ?? 'claude',
- cwd: res.worktreePath ?? cwd,
- command,
- role: meta.role,
- worktreePath: res.worktreePath
- });
- } catch { /* window torn down — the worker still runs headless */ }
-
// Register for done-scan / idle-reap / token-cap / safe teardown (pty id == workerId).
// tokenCap is optional plumbing (default unlimited) — only a positive finite cap is kept.
const tokenCap = typeof raw.tokenCap === 'number' && Number.isFinite(raw.tokenCap) && raw.tokenCap > 0
@@ -4588,21 +4531,8 @@ async function ephemeralWorkerTick(): Promise<void> {
const defaultTokenCap = typeof cfg.defaultWorkerTokenCap === 'number' && cfg.defaultWorkerTokenCap > 0
? cfg.defaultWorkerTokenCap : 0;
- // (1) Finish or reap. `releasing` guards the gap before teardown lands.
- // ptyManager.kill(workerId) alone is NOT enough here (D10): kill() deletes
- // its PTY session from the manager's map SYNCHRONOUSLY, but the process's
- // real exit — and node-pty's onExit, which is what fires the exitHandler
- // that calls teardownPty — only arrives ASYNCHRONOUSLY afterward. By the
- // time it does, the map lookup that onExit uses to detect "was this id
- // reclaimed by a respawn" already reads empty, so its stale-exit guard
- // misfires and the exitHandler is skipped every time. Confirmed live: a
- // reaped worker's process was gone (`ps` had nothing) while registry.json
- // still read `archived: false` and fleet.json still listed it — which
- // means the LIVE ROSTER god's own prompt is built from (rosterContext(),
- // fed by fleet.json) kept instructing "route work to" a dead process,
- // forever, after every reap. teardownPty is idempotent (guarded on map
- // presence), so calling it explicitly here is safe even on the rare tick
- // where the async path does still land.
+ // (1) Finish or reap. ptyManager.kill → teardownPty → gated worktree + archive
+ // + liveWorkers.delete. `releasing` guards the gap before onExit fires.
for (const [workerId, rec] of [...liveWorkers]) {
if (rec.releasing) continue;
if (workerSignaledDone(workerId, rec.spawnedAt)) {
@@ -4610,7 +4540,6 @@ async function ephemeralWorkerTick(): Promise<void> {
rec.releasing = true;
console.log(`[worker] ${workerId} signaled done — releasing`);
ptyManager.kill(workerId);
- teardownPty(workerId);
continue;
}
// Token-cap reap (default-off plumbing). An effective cap > 0 → reap when the
@@ -4627,7 +4556,6 @@ async function ephemeralWorkerTick(): Promise<void> {
rec.slack
);
ptyManager.kill(workerId);
- teardownPty(workerId);
continue;
}
}
@@ -4642,7 +4570,6 @@ async function ephemeralWorkerTick(): Promise<void> {
rec.slack
);
ptyManager.kill(workerId);
- teardownPty(workerId);
}
}