@openape/nest 1.0.0 → 1.1.1

package/dist/index.mjs

@@ -95,19 +95,18 @@ async function handleAgentSpawn(ctx) {
     } : void 0
   };
   upsertAgent(entry);
-  ctx.supervisor.reconcile(listAgents());
+  await ctx.supervisor.reconcile(listAgents());
   return { name, email: entry.email, uid, home: entry.home };
 }
 async function handleAgentDestroy(ctx, name) {
   if (!NAME_REGEX.test(name)) throw new Error(`invalid agent name "${name}"`);
   const entry = findAgent(name);
   if (!entry) throw new Error(`agent "${name}" not registered with this nest`);
-  ctx.supervisor.stop(name);
   ctx.log(`nest: destroying agent "${name}"...`);
   const args = ["run", "--as", "root", "--", "apes", "agents", "destroy", name, "--force"];
   await execFileAsync(ctx.apesBin, args, { maxBuffer: 4 * 1024 * 1024 });
   removeAgent(name);
-  ctx.supervisor.reconcile(listAgents());
+  await ctx.supervisor.reconcile(listAgents());
   return { name, removed: true };
 }
 async function readUidFromDscl(name) {
@@ -233,109 +232,98 @@ function reapStaleResponses(log2) {
   }
 }
 
-// src/lib/supervisor.ts
-import { spawn } from "child_process";
+// src/lib/pm2-supervisor.ts
+import { execFile as execFile2 } from "child_process";
+import { mkdirSync as mkdirSync3, writeFileSync as writeFileSync3 } from "fs";
+import { join as join3 } from "path";
 import process from "process";
-var MIN_BACKOFF_MS = 2e3;
-var MAX_BACKOFF_MS = 6e4;
-var STABLE_RUNTIME_MS = 3e4;
-var Supervisor = class {
+import { promisify as promisify2 } from "util";
+var execFileAsync2 = promisify2(execFile2);
+var AGENTS_DIR = "/var/openape/agents";
+function pm2AppName(agentName) {
+  return `openape-bridge-${agentName}`;
+}
+function ecosystemPath(agentName) {
+  return join3(AGENTS_DIR, agentName, "ecosystem.config.js");
+}
+function ecosystemContents(apesBin, agentName) {
+  void apesBin;
+  return `// Auto-generated by Pm2Supervisor for agent '${agentName}'.
+// Edit at runtime via:
+//   apes run --as ${agentName} -- pm2 reload ${pm2AppName(agentName)}
+module.exports = {
+  apps: [{
+    name: '${pm2AppName(agentName)}',
+    script: 'openape-chat-bridge',
+    autorestart: true,
+    max_restarts: 10,
+    min_uptime: '30s',
+    restart_delay: 2000,
+    merge_logs: true,
+  }],
+}
+`;
+}
+var Pm2Supervisor = class {
   constructor(deps) {
     this.deps = deps;
   }
-  children = /* @__PURE__ */ new Map();
-  /** Bring the supervised set in line with the desired set. */
-  reconcile(desired) {
-    const desiredNames = new Set(desired.filter((a) => a.bridge != null).map((a) => a.name));
-    for (const [name] of this.children) {
-      if (!desiredNames.has(name)) this.stop(name);
-    }
+  inflight = /* @__PURE__ */ new Set();
+  /** Bring per-agent pm2 state in line with the registry. Idempotent. */
+  async reconcile(desired) {
     for (const agent of desired) {
       if (agent.bridge == null) continue;
-      if (!this.children.has(agent.name)) this.start(agent);
+      if (this.inflight.has(agent.name)) continue;
+      this.inflight.add(agent.name);
+      try {
+        await this.startOrReload(agent.name);
+      } catch (err) {
+        this.deps.log(`pm2-supervisor: ${agent.name} startOrReload failed: ${err instanceof Error ? err.message.split("\n")[0] : String(err)}`);
+      } finally {
+        this.inflight.delete(agent.name);
+      }
     }
   }
-  size() {
-    return this.children.size;
-  }
-  status() {
-    const now = Date.now();
-    return Array.from(this.children.entries()).map(([name, s]) => ({
-      name,
-      pid: s.child.pid ?? -1,
-      uptimeSec: Math.floor((now - s.startedAt) / 1e3),
-      consecutiveCrashes: s.consecutiveCrashes
-    }));
-  }
-  start(agent) {
-    if (this.children.has(agent.name)) return;
-    this.deps.log(`supervisor: starting bridge for ${agent.name}`);
-    this.spawnChild(agent, 0);
-  }
-  stop(name) {
-    const s = this.children.get(name);
-    if (!s) return;
-    this.deps.log(`supervisor: stopping bridge for ${name}`);
-    if (s.restartTimer) clearTimeout(s.restartTimer);
-    this.children.delete(name);
+  /** Stop one agent's pm2 app — used at destroy time. */
+  async stop(agentName) {
+    const name = pm2AppName(agentName);
     try {
-      s.child.kill("SIGTERM");
-    } catch {
+      await this.runAsAgent(agentName, ["pm2", "delete", name]);
+      this.deps.log(`pm2-supervisor: deleted ${name}`);
+    } catch (err) {
+      this.deps.log(`pm2-supervisor: delete ${name}: ${err instanceof Error ? err.message.split("\n")[0] : String(err)}`);
     }
   }
-  stopAll() {
-    for (const name of Array.from(this.children.keys())) this.stop(name);
+  /** Best-effort cleanup — called on Nest shutdown. We don't kill
+   *  the per-agent pm2-daemons; they should keep running so bridges
+   *  stay alive across Nest restarts. No-op for now. */
+  async stopAll() {
   }
-  spawnChild(agent, prevCrashes) {
-    const args = ["run", "--as", agent.name, "--wait", "--", "openape-chat-bridge"];
-    const child = spawn(this.deps.apesBin, args, {
-      stdio: ["ignore", "pipe", "pipe"],
-      detached: false,
-      // Inherit env — most importantly PATH (host bin dirs from
-      // captureHostBinDirs at install time) and HOME (the Nest's
-      // data dir, where its own auth.json lives so apes-cli reads
-      // the nest identity for the YOLO grant).
-      env: process.env
-    });
-    child.stdout?.on("data", (chunk) => this.forwardLog(agent.name, "stdout", chunk));
-    child.stderr?.on("data", (chunk) => this.forwardLog(agent.name, "stderr", chunk));
-    const supervised = {
-      child,
-      consecutiveCrashes: prevCrashes,
-      startedAt: Date.now()
-    };
-    this.children.set(agent.name, supervised);
-    child.on("exit", (code, signal) => {
-      const stillManaged = this.children.get(agent.name) === supervised;
-      if (!stillManaged) return;
-      const ranLongEnough = Date.now() - supervised.startedAt > STABLE_RUNTIME_MS;
-      const nextCrashes = ranLongEnough ? 1 : prevCrashes + 1;
-      const backoff = Math.min(MAX_BACKOFF_MS, MIN_BACKOFF_MS * 2 ** Math.max(0, nextCrashes - 1));
-      this.deps.log(
-        `supervisor: ${agent.name} bridge exited code=${code} signal=${signal ?? "none"} consecutive=${nextCrashes} \u2192 respawn in ${backoff}ms`
-      );
-      supervised.restartTimer = setTimeout(() => {
-        if (this.children.get(agent.name) !== supervised) return;
-        this.children.delete(agent.name);
-        this.spawnChild(agent, nextCrashes);
-      }, backoff);
-    });
+  async startOrReload(agentName) {
+    mkdirSync3(AGENTS_DIR, { recursive: true, mode: 493 });
+    const dir = join3(AGENTS_DIR, agentName);
+    mkdirSync3(dir, { recursive: true, mode: 493 });
+    const path = ecosystemPath(agentName);
+    writeFileSync3(path, ecosystemContents(this.deps.apesBin, agentName), { mode: 420 });
+    await this.runAsAgent(agentName, ["pm2", "startOrReload", path]);
+    this.deps.log(`pm2-supervisor: ${agentName} bridge (re)started via pm2`);
   }
-  forwardLog(name, stream, chunk) {
-    const text = chunk.toString("utf8");
-    for (const line of text.split("\n")) {
-      const trimmed = line.trimEnd();
-      if (!trimmed) continue;
-      this.deps.log(`[${name}/${stream}] ${trimmed}`);
-    }
+  /** Run a pm2 subcommand AS the agent — escapes-helper does the
+   *  setuid switch, then exec's pm2 in the agent's uid. */
+  async runAsAgent(agentName, args) {
+    return execFileAsync2(
+      this.deps.apesBin,
+      ["run", "--as", agentName, "--wait", "--", ...args],
+      { maxBuffer: 1024 * 1024, env: process.env, timeout: 6e4 }
+    );
   }
 };
 
 // src/lib/troop-sync.ts
-import { execFile as execFile2 } from "child_process";
+import { execFile as execFile3 } from "child_process";
 import process2 from "process";
-import { promisify as promisify2 } from "util";
-var execFileAsync2 = promisify2(execFile2);
+import { promisify as promisify3 } from "util";
+var execFileAsync3 = promisify3(execFile3);
 var TICK_MS = 5 * 60 * 1e3;
 var TroopSync = class {
   constructor(deps) {
@@ -369,7 +357,7 @@ var TroopSync = class {
   }
   async syncOne(name) {
     try {
-      await execFileAsync2(
+      await execFileAsync3(
         this.deps.apesBin,
         ["run", "--as", name, "--wait", "--", "apes", "agents", "sync"],
         { maxBuffer: 1024 * 1024, env: process2.env, timeout: 6e4 }
@@ -386,17 +374,20 @@ function log(line) {
   process3.stderr.write(`${(/* @__PURE__ */ new Date()).toISOString()}  ${line}
 `);
 }
-var supervisor = new Supervisor({ apesBin: APES_BIN, log });
+var supervisor = new Pm2Supervisor({ apesBin: APES_BIN, log });
 var troopSync = new TroopSync({ apesBin: APES_BIN, log });
 var intentChannel = new IntentChannel({ apesBin: APES_BIN, supervisor, log });
-supervisor.reconcile(listAgents());
-log(`nest: supervisor reconciled, ${supervisor.size()} bridge process(es) starting`);
+void supervisor.reconcile(listAgents()).then(
+  () => log("nest: pm2-supervisor reconciled with registry")
+).catch((err) => {
+  log(`nest: pm2-supervisor reconcile failed: ${err instanceof Error ? err.message : String(err)}`);
+});
 troopSync.start();
 intentChannel.start();
 var reaperTimer = setInterval(reapStaleResponses, 60 * 60 * 1e3, log);
 process3.on("SIGTERM", () => {
   log("nest: SIGTERM \u2014 stopping");
-  supervisor.stopAll();
+  void supervisor.stopAll();
   troopSync.stop();
   intentChannel.stop();
   clearInterval(reaperTimer);
@@ -404,7 +395,7 @@ process3.on("SIGTERM", () => {
 });
 process3.on("SIGINT", () => {
   log("nest: SIGINT \u2014 stopping");
-  supervisor.stopAll();
+  void supervisor.stopAll();
   troopSync.stop();
   intentChannel.stop();
   clearInterval(reaperTimer);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openape/nest",
-  "version": "1.0.0",
+  "version": "1.1.1",
   "description": "OpenApe Nest — local control-plane daemon that supervises agent processes on this computer. Talks to troop SP for ownership state, spawns/destroys agents via DDISA always-grants, supervises chat-bridge children (replacing per-agent launchd plists).",
   "type": "module",
   "license": "MIT",