Add verae-keep: host supervisor with watch+guard on NS1.

Restarts crashed units unless the admin console paused or stopped
them. Fleet writes per-instance intent.json. Tested on 138:
crash-restart, pause-hold, watch respawn of keep.

packages/verae-fleet/src/ssh.js

@@ -135,6 +135,21 @@ export async function sshSpawnWorker(machine, { instance, env }) {
   return { pid, remoteDir };
 }
 
+export async function sshWriteIntent(machine, instance, state) {
+  const remoteDir = machine.remoteDir || '~/verae-fleet-runtime';
+  const body = JSON.stringify({
+    id: instance,
+    state,
+    source: 'fleet',
+    updatedAt: new Date().toISOString(),
+  });
+  const dest = `${remoteDir}/data/${instance}/intent.json`;
+  await sshExec(
+    machine,
+    `mkdir -p ${remoteDir}/data/${instance} && printf %s ${shellQuote(body)} > ${dest}`,
+  );
+}
+
 export async function sshKillWorker(machine, pid, instance) {
   const cmd = pid
     ? `kill -TERM ${Number(pid)} 2>/dev/null || true`

packages/verae-fleet/src/supervisor.js

@@ -12,7 +12,7 @@ import { httpProbe, postControl } from './health.js';
 import { classifyService } from './classify.js';
 import { pickMachine, upsertMachine, saveOverlay } from './machines.js';
 import { summarizeRtt } from './rtt.js';
-import { sshSpawnWorker, sshKillWorker, sshHttp, sshCheck } from './ssh.js';
+import { sshSpawnWorker, sshKillWorker, sshHttp, sshCheck, sshWriteIntent } from './ssh.js';
 
 const WORKER = path.join(path.dirname(fileURLToPath(import.meta.url)), 'worker.js');
 
@@ -237,13 +237,41 @@ export class Supervisor {
     const h = await waitForHealth(rec, 8000);
     rec.healthy = h.ok;
     rec.lastProbe = h;
+    await this.persistIntent(rec, 'run');
     return rec;
   }
 
+  async persistIntent(rec, state) {
+    if (!rec) return;
+    const body = {
+      id: rec.id,
+      state,
+      source: 'fleet',
+      updatedAt: new Date().toISOString(),
+    };
+    try {
+      fs.mkdirSync(rec.stateDir, { recursive: true });
+      fs.writeFileSync(path.join(rec.stateDir, 'intent.json'), JSON.stringify(body, null, 2));
+    } catch {
+      /* ignore */
+    }
+    if (rec.kind === 'ssh') {
+      const m = this.machines().find((x) => x.id === rec.machine);
+      if (m) {
+        try {
+          await sshWriteIntent(m, rec.id, state);
+        } catch {
+          /* ignore */
+        }
+      }
+    }
+  }
+
   async stopInstance(instanceId, { replace = true } = {}) {
     const rec = this.instances.get(instanceId);
     if (!rec) return null;
     rec.stopping = true;
+    await this.persistIntent(rec, 'stop');
     if (rec.kind === 'ssh') {
       const m = this.machines().find((x) => x.id === rec.machine);
       if (m) {
@@ -277,6 +305,7 @@ export class Supervisor {
     if (!rec) throw new Error(`no instance ${instanceId}`);
     await this.controlInstance(rec, '/pause');
     rec.paused = true;
+    await this.persistIntent(rec, 'pause');
     this.log('pause', { instance: instanceId, service: rec.service });
     if (replace) await this.reconcile(rec.service);
     return rec;
@@ -287,6 +316,7 @@ export class Supervisor {
     if (!rec) throw new Error(`no instance ${instanceId}`);
     await this.controlInstance(rec, '/resume');
     rec.paused = false;
+    await this.persistIntent(rec, 'run');
     this.log('resume', { instance: instanceId, service: rec.service });
     return rec;
   }