agentflow-core 0.2.3 → 0.3.1

package/dist/cli.cjs

@@ -1109,6 +1109,522 @@ function startLive(argv) {
   });
 }
 
+// src/watch.ts
+var import_node_fs4 = require("fs");
+var import_node_path3 = require("path");
+var import_node_os = require("os");
+
+// src/watch-state.ts
+var import_node_fs3 = require("fs");
+function parseDuration(input) {
+  const match = input.match(/^(\d+(?:\.\d+)?)\s*(s|m|h|d)$/i);
+  if (!match) {
+    const n = parseInt(input, 10);
+    return isNaN(n) ? 0 : n * 1e3;
+  }
+  const value = parseFloat(match[1]);
+  switch (match[2].toLowerCase()) {
+    case "s":
+      return value * 1e3;
+    case "m":
+      return value * 6e4;
+    case "h":
+      return value * 36e5;
+    case "d":
+      return value * 864e5;
+    default:
+      return value * 1e3;
+  }
+}
+function emptyState() {
+  return { version: 1, agents: {}, lastPollTime: 0 };
+}
+function loadWatchState(filePath) {
+  if (!(0, import_node_fs3.existsSync)(filePath)) return emptyState();
+  try {
+    const raw = JSON.parse((0, import_node_fs3.readFileSync)(filePath, "utf8"));
+    if (raw.version !== 1 || typeof raw.agents !== "object") return emptyState();
+    return raw;
+  } catch {
+    return emptyState();
+  }
+}
+function saveWatchState(filePath, state) {
+  const tmp = filePath + ".tmp";
+  try {
+    (0, import_node_fs3.writeFileSync)(tmp, JSON.stringify(state, null, 2), "utf8");
+    (0, import_node_fs3.renameSync)(tmp, filePath);
+  } catch {
+    try {
+      (0, import_node_fs3.writeFileSync)(filePath, JSON.stringify(state, null, 2), "utf8");
+    } catch {
+    }
+  }
+}
+function estimateInterval(history) {
+  if (history.length < 3) return 0;
+  const sorted = [...history].sort((a, b) => a - b);
+  const deltas = [];
+  for (let i = 1; i < sorted.length; i++) {
+    const d = sorted[i] - sorted[i - 1];
+    if (d > 0) deltas.push(d);
+  }
+  if (deltas.length === 0) return 0;
+  deltas.sort((a, b) => a - b);
+  return deltas[Math.floor(deltas.length / 2)];
+}
+function detectTransitions(previous, currentRecords, config, now) {
+  const alerts = [];
+  const hasError = config.alertConditions.some((c) => c.type === "error");
+  const hasRecovery = config.alertConditions.some((c) => c.type === "recovery");
+  const staleConditions = config.alertConditions.filter((c) => c.type === "stale");
+  const consecutiveConditions = config.alertConditions.filter((c) => c.type === "consecutive-errors");
+  const byAgent = /* @__PURE__ */ new Map();
+  for (const r of currentRecords) {
+    const existing = byAgent.get(r.id);
+    if (!existing || r.lastActive > existing.lastActive) {
+      byAgent.set(r.id, r);
+    }
+  }
+  for (const [agentId, record] of byAgent) {
+    const prev = previous.agents[agentId];
+    const prevStatus = prev?.lastStatus ?? "unknown";
+    const currStatus = record.status;
+    if (hasError && currStatus === "error" && prevStatus !== "error") {
+      if (canAlert(prev, "error", config.cooldownMs, now)) {
+        alerts.push(makePayload(agentId, "error", prevStatus, currStatus, record, config.dirs));
+      }
+    }
+    if (hasRecovery && currStatus === "ok" && prevStatus === "error") {
+      alerts.push(makePayload(agentId, "recovery", prevStatus, currStatus, record, config.dirs));
+    }
+    const newConsec = currStatus === "error" ? (prev?.consecutiveErrors ?? 0) + 1 : 0;
+    for (const cond of consecutiveConditions) {
+      if (newConsec === cond.threshold) {
+        if (canAlert(prev, `consecutive-errors:${cond.threshold}`, config.cooldownMs, now)) {
+          alerts.push(makePayload(
+            agentId,
+            `consecutive-errors (${cond.threshold})`,
+            prevStatus,
+            currStatus,
+            { ...record, detail: `${newConsec} consecutive errors. ${record.detail}` },
+            config.dirs
+          ));
+        }
+      }
+    }
+    for (const cond of staleConditions) {
+      const sinceActive = now - record.lastActive;
+      if (sinceActive > cond.durationMs && record.lastActive > 0) {
+        if (canAlert(prev, "stale", config.cooldownMs, now)) {
+          const mins = Math.floor(sinceActive / 6e4);
+          alerts.push(makePayload(
+            agentId,
+            "stale",
+            prevStatus,
+            currStatus,
+            { ...record, detail: `No update for ${mins}m. ${record.detail}` },
+            config.dirs
+          ));
+        }
+      }
+    }
+    if (staleConditions.length === 0) {
+      const history = prev?.mtimeHistory ?? [];
+      const expectedInterval = estimateInterval(history);
+      if (expectedInterval > 0) {
+        const sinceActive = now - record.lastActive;
+        if (sinceActive > expectedInterval * 3) {
+          if (canAlert(prev, "stale-auto", config.cooldownMs, now)) {
+            const mins = Math.floor(sinceActive / 6e4);
+            const expectedMins = Math.floor(expectedInterval / 6e4);
+            alerts.push(makePayload(
+              agentId,
+              "stale (auto)",
+              prevStatus,
+              currStatus,
+              { ...record, detail: `No update for ${mins}m (expected every ~${expectedMins}m). ${record.detail}` },
+              config.dirs
+            ));
+          }
+        }
+      }
+    }
+  }
+  return alerts;
+}
+function updateWatchState(state, records, alerts, now) {
+  const agents = { ...state.agents };
+  const alertsByAgent = /* @__PURE__ */ new Map();
+  for (const a of alerts) alertsByAgent.set(a.agentId, a);
+  const byAgent = /* @__PURE__ */ new Map();
+  for (const r of records) {
+    const existing = byAgent.get(r.id);
+    if (!existing || r.lastActive > existing.lastActive) {
+      byAgent.set(r.id, r);
+    }
+  }
+  for (const [agentId, record] of byAgent) {
+    const prev = agents[agentId];
+    const history = prev?.mtimeHistory ?? [];
+    const newHistory = [...history];
+    if (newHistory.length === 0 || newHistory[newHistory.length - 1] !== record.lastActive) {
+      newHistory.push(record.lastActive);
+    }
+    while (newHistory.length > 10) newHistory.shift();
+    const alert = alertsByAgent.get(agentId);
+    const consecutiveErrors = record.status === "error" ? (prev?.consecutiveErrors ?? 0) + 1 : 0;
+    agents[agentId] = {
+      id: agentId,
+      lastStatus: record.status,
+      lastActive: record.lastActive,
+      lastAlertTime: alert ? now : prev?.lastAlertTime ?? 0,
+      lastAlertReason: alert ? alert.condition : prev?.lastAlertReason ?? "",
+      consecutiveErrors,
+      mtimeHistory: newHistory
+    };
+  }
+  return { version: 1, agents, lastPollTime: now };
+}
+function canAlert(prev, reason, cooldownMs, now) {
+  if (!prev) return true;
+  if (prev.lastAlertReason !== reason) return true;
+  return now - prev.lastAlertTime > cooldownMs;
+}
+function makePayload(agentId, condition, previousStatus, currentStatus, record, dirs) {
+  return {
+    agentId,
+    condition,
+    previousStatus,
+    currentStatus,
+    detail: record.detail,
+    file: record.file,
+    timestamp: Date.now(),
+    dirs
+  };
+}
+
+// src/watch-alerts.ts
+var import_node_https = require("https");
+var import_node_http = require("http");
+var import_node_child_process2 = require("child_process");
+function formatAlertMessage(payload) {
+  const time = new Date(payload.timestamp).toISOString();
+  const arrow = `${payload.previousStatus} \u2192 ${payload.currentStatus}`;
+  return [
+    `[ALERT] ${payload.condition}: "${payload.agentId}"`,
+    `  Status:  ${arrow}`,
+    payload.detail ? `  Detail:  ${payload.detail}` : null,
+    `  File:    ${payload.file}`,
+    `  Time:    ${time}`
+  ].filter(Boolean).join("\n");
+}
+function formatTelegram(payload) {
+  const icon = payload.condition === "recovery" ? "\u2705" : "\u26A0\uFE0F";
+  const time = new Date(payload.timestamp).toLocaleTimeString();
+  return [
+    `${icon} *AgentFlow Alert*`,
+    `*${payload.condition}*: \`${payload.agentId}\``,
+    `Status: ${payload.previousStatus} \u2192 ${payload.currentStatus}`,
+    payload.detail ? `Detail: ${payload.detail.slice(0, 200)}` : null,
+    `Time: ${time}`
+  ].filter(Boolean).join("\n");
+}
+async function sendAlert(payload, channel) {
+  try {
+    switch (channel.type) {
+      case "stdout":
+        sendStdout(payload);
+        break;
+      case "telegram":
+        await sendTelegram(payload, channel.botToken, channel.chatId);
+        break;
+      case "webhook":
+        await sendWebhook(payload, channel.url);
+        break;
+      case "command":
+        await sendCommand(payload, channel.cmd);
+        break;
+    }
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err);
+    console.error(`[agentflow] Failed to send ${channel.type} alert: ${msg}`);
+  }
+}
+function sendStdout(payload) {
+  console.log(formatAlertMessage(payload));
+}
+function sendTelegram(payload, botToken, chatId) {
+  const body = JSON.stringify({
+    chat_id: chatId,
+    text: formatTelegram(payload),
+    parse_mode: "Markdown"
+  });
+  return new Promise((resolve5, reject) => {
+    const req = (0, import_node_https.request)(
+      `https://api.telegram.org/bot${botToken}/sendMessage`,
+      { method: "POST", headers: { "Content-Type": "application/json", "Content-Length": Buffer.byteLength(body) } },
+      (res) => {
+        res.resume();
+        if (res.statusCode && res.statusCode >= 200 && res.statusCode < 300) resolve5();
+        else reject(new Error(`Telegram API returned ${res.statusCode}`));
+      }
+    );
+    req.on("error", reject);
+    req.write(body);
+    req.end();
+  });
+}
+function sendWebhook(payload, url) {
+  const body = JSON.stringify(payload);
+  const isHttps = url.startsWith("https");
+  const doRequest = isHttps ? import_node_https.request : import_node_http.request;
+  return new Promise((resolve5, reject) => {
+    const req = doRequest(
+      url,
+      { method: "POST", headers: { "Content-Type": "application/json", "Content-Length": Buffer.byteLength(body) } },
+      (res) => {
+        res.resume();
+        if (res.statusCode && res.statusCode >= 200 && res.statusCode < 300) resolve5();
+        else reject(new Error(`Webhook returned ${res.statusCode}`));
+      }
+    );
+    req.on("error", reject);
+    req.setTimeout(1e4, () => {
+      req.destroy(new Error("Webhook timeout"));
+    });
+    req.write(body);
+    req.end();
+  });
+}
+function sendCommand(payload, cmd) {
+  return new Promise((resolve5, reject) => {
+    const env = {
+      ...process.env,
+      AGENTFLOW_ALERT_AGENT: payload.agentId,
+      AGENTFLOW_ALERT_CONDITION: payload.condition,
+      AGENTFLOW_ALERT_STATUS: payload.currentStatus,
+      AGENTFLOW_ALERT_PREVIOUS_STATUS: payload.previousStatus,
+      AGENTFLOW_ALERT_DETAIL: payload.detail,
+      AGENTFLOW_ALERT_FILE: payload.file,
+      AGENTFLOW_ALERT_TIMESTAMP: String(payload.timestamp)
+    };
+    (0, import_node_child_process2.exec)(cmd, { env, timeout: 3e4 }, (err) => {
+      if (err) reject(err);
+      else resolve5();
+    });
+  });
+}
+
+// src/watch.ts
+function parseWatchArgs(argv) {
+  const dirs = [];
+  const alertConditions = [];
+  const notifyChannels = [];
+  let recursive = false;
+  let pollIntervalMs = 3e4;
+  let cooldownMs = 30 * 6e4;
+  let stateFilePath = "";
+  const args = argv.slice(0);
+  if (args[0] === "watch") args.shift();
+  let i = 0;
+  while (i < args.length) {
+    const arg = args[i];
+    if (arg === "--help" || arg === "-h") {
+      printWatchUsage();
+      process.exit(0);
+    } else if (arg === "--alert-on") {
+      i++;
+      const val = args[i] ?? "";
+      if (val === "error") {
+        alertConditions.push({ type: "error" });
+      } else if (val === "recovery") {
+        alertConditions.push({ type: "recovery" });
+      } else if (val.startsWith("stale:")) {
+        const dur = parseDuration(val.slice(6));
+        if (dur > 0) alertConditions.push({ type: "stale", durationMs: dur });
+      } else if (val.startsWith("consecutive-errors:")) {
+        const n = parseInt(val.slice(19), 10);
+        if (n > 0) alertConditions.push({ type: "consecutive-errors", threshold: n });
+      }
+      i++;
+    } else if (arg === "--notify") {
+      i++;
+      const val = args[i] ?? "";
+      if (val === "telegram") {
+        const botToken = process.env["AGENTFLOW_TELEGRAM_BOT_TOKEN"] ?? "";
+        const chatId = process.env["AGENTFLOW_TELEGRAM_CHAT_ID"] ?? "";
+        if (botToken && chatId) {
+          notifyChannels.push({ type: "telegram", botToken, chatId });
+        } else {
+          console.error("Warning: --notify telegram requires AGENTFLOW_TELEGRAM_BOT_TOKEN and AGENTFLOW_TELEGRAM_CHAT_ID env vars");
+        }
+      } else if (val.startsWith("webhook:")) {
+        notifyChannels.push({ type: "webhook", url: val.slice(8) });
+      } else if (val.startsWith("command:")) {
+        notifyChannels.push({ type: "command", cmd: val.slice(8) });
+      }
+      i++;
+    } else if (arg === "--poll") {
+      i++;
+      const v = parseInt(args[i] ?? "", 10);
+      if (!isNaN(v) && v > 0) pollIntervalMs = v * 1e3;
+      i++;
+    } else if (arg === "--cooldown") {
+      i++;
+      const dur = parseDuration(args[i] ?? "30m");
+      if (dur > 0) cooldownMs = dur;
+      i++;
+    } else if (arg === "--state-file") {
+      i++;
+      stateFilePath = args[i] ?? "";
+      i++;
+    } else if (arg === "--recursive" || arg === "-R") {
+      recursive = true;
+      i++;
+    } else if (!arg.startsWith("-")) {
+      dirs.push((0, import_node_path3.resolve)(arg));
+      i++;
+    } else {
+      i++;
+    }
+  }
+  if (dirs.length === 0) dirs.push((0, import_node_path3.resolve)("."));
+  if (alertConditions.length === 0) {
+    alertConditions.push({ type: "error" });
+    alertConditions.push({ type: "recovery" });
+  }
+  notifyChannels.unshift({ type: "stdout" });
+  if (!stateFilePath) {
+    stateFilePath = (0, import_node_path3.join)(dirs[0], ".agentflow-watch-state.json");
+  }
+  return {
+    dirs,
+    recursive,
+    pollIntervalMs,
+    alertConditions,
+    notifyChannels,
+    stateFilePath: (0, import_node_path3.resolve)(stateFilePath),
+    cooldownMs
+  };
+}
+function printWatchUsage() {
+  console.log(`
+AgentFlow Watch \u2014 headless alert system for agent infrastructure.
+
+Polls directories for JSON/JSONL files, detects failures and stale
+agents, sends alerts. Same auto-detection as \`agentflow live\`.
+
+Usage:
+  agentflow watch [dir...] [options]
+
+Arguments:
+  dir                          One or more directories to watch (default: .)
+
+Alert conditions (--alert-on, repeatable):
+  error                        Agent transitions to error status
+  recovery                     Agent recovers from error to ok
+  stale:DURATION               No file update within duration (e.g. 15m, 1h)
+  consecutive-errors:N         N consecutive error observations
+
+  Default (if none specified): error + recovery
+
+Notification channels (--notify, repeatable):
+  telegram                     Telegram Bot API (needs env vars)
+  webhook:URL                  POST JSON to any URL
+  command:CMD                  Run shell command with alert env vars
+
+  Stdout alerts are always printed regardless of --notify flags.
+
+Options:
+  --poll <secs>                Poll interval in seconds (default: 30)
+  --cooldown <duration>        Alert dedup cooldown (default: 30m)
+  --state-file <path>          Persistence file (default: <dir>/.agentflow-watch-state.json)
+  -R, --recursive              Scan subdirectories (1 level deep)
+  -h, --help                   Show this help message
+
+Environment variables:
+  AGENTFLOW_TELEGRAM_BOT_TOKEN   Telegram bot token (for --notify telegram)
+  AGENTFLOW_TELEGRAM_CHAT_ID     Telegram chat ID (for --notify telegram)
+
+Examples:
+  agentflow watch ./data --alert-on error --alert-on stale:15m
+  agentflow watch ./data ./cron --notify telegram --poll 60
+  agentflow watch ./traces --notify webhook:https://hooks.slack.com/... --alert-on consecutive-errors:3
+  agentflow watch ./data --notify "command:curl -X POST https://my-pagerduty/alert"
+`.trim());
+}
+function startWatch(argv) {
+  const config = parseWatchArgs(argv);
+  const valid = config.dirs.filter((d) => (0, import_node_fs4.existsSync)(d));
+  if (valid.length === 0) {
+    console.error(`No valid directories found: ${config.dirs.join(", ")}`);
+    process.exit(1);
+  }
+  const invalid = config.dirs.filter((d) => !(0, import_node_fs4.existsSync)(d));
+  if (invalid.length > 0) {
+    console.warn(`Skipping non-existent: ${invalid.join(", ")}`);
+  }
+  let state = loadWatchState(config.stateFilePath);
+  const condLabels = config.alertConditions.map((c) => {
+    if (c.type === "stale") return `stale:${Math.floor(c.durationMs / 6e4)}m`;
+    if (c.type === "consecutive-errors") return `consecutive-errors:${c.threshold}`;
+    return c.type;
+  });
+  const channelLabels = config.notifyChannels.filter((c) => c.type !== "stdout").map((c) => {
+    if (c.type === "webhook") return `webhook:${c.url.slice(0, 40)}...`;
+    if (c.type === "command") return `command:${c.cmd.slice(0, 40)}`;
+    return c.type;
+  });
+  console.log(`
+agentflow watch started`);
+  console.log(`  Directories:  ${valid.join(", ")}`);
+  console.log(`  Poll:         ${config.pollIntervalMs / 1e3}s`);
+  console.log(`  Alert on:     ${condLabels.join(", ")}`);
+  console.log(`  Notify:       stdout${channelLabels.length > 0 ? ", " + channelLabels.join(", ") : ""}`);
+  console.log(`  Cooldown:     ${Math.floor(config.cooldownMs / 6e4)}m`);
+  console.log(`  State:        ${config.stateFilePath}`);
+  console.log(`  Hostname:     ${(0, import_node_os.hostname)()}`);
+  console.log("");
+  let pollCount = 0;
+  async function poll() {
+    const now = Date.now();
+    pollCount++;
+    const files = scanFiles(valid, config.recursive);
+    const records = [];
+    for (const f of files.slice(0, 500)) {
+      const recs = f.ext === ".jsonl" ? processJsonlFile(f) : processJsonFile(f);
+      records.push(...recs);
+    }
+    const alerts = detectTransitions(state, records, config, now);
+    for (const alert of alerts) {
+      for (const channel of config.notifyChannels) {
+        await sendAlert(alert, channel);
+      }
+    }
+    state = updateWatchState(state, records, alerts, now);
+    saveWatchState(config.stateFilePath, state);
+    if (pollCount % 10 === 0) {
+      const agentCount = Object.keys(state.agents).length;
+      const errorCount = Object.values(state.agents).filter((a) => a.lastStatus === "error").length;
+      const runningCount = Object.values(state.agents).filter((a) => a.lastStatus === "running").length;
+      const time = (/* @__PURE__ */ new Date()).toLocaleTimeString();
+      console.log(`[${time}] heartbeat: ${agentCount} agents, ${runningCount} running, ${errorCount} errors, ${files.length} files`);
+    }
+  }
+  poll();
+  setInterval(() => {
+    poll();
+  }, config.pollIntervalMs);
+  function shutdown() {
+    console.log("\nagentflow watch stopped.");
+    saveWatchState(config.stateFilePath, state);
+    process.exit(0);
+  }
+  process.on("SIGINT", shutdown);
+  process.on("SIGTERM", shutdown);
+}
+
 // src/cli.ts
 function printHelp() {
   console.log(`
@@ -1118,8 +1634,9 @@ Usage:
   agentflow <command> [options]
 
 Commands:
-  run   [options] -- <cmd>       Wrap a command with automatic execution tracing
-  live  [dir...] [options]      Real-time terminal monitor (auto-detects any JSON/JSONL)
+  run    [options] -- <cmd>       Wrap a command with automatic execution tracing
+  live   [dir...] [options]      Real-time terminal monitor (auto-detects any JSON/JSONL)
+  watch  [dir...] [options]      Headless alert system \u2014 detects failures, sends notifications
 
 Run \`agentflow <command> --help\` for command-specific options.
 
@@ -1127,7 +1644,8 @@ Examples:
   agentflow run --traces-dir ./traces -- python -m myagent process
   agentflow live ./data
   agentflow live ./traces ./cron ./workers -R
-  agentflow live ./data --refresh 5
+  agentflow watch ./data --alert-on error --notify telegram
+  agentflow watch ./data ./cron --alert-on stale:15m --notify webhook:https://...
 `.trim());
 }
 function parseRunArgs(argv) {
@@ -1265,7 +1783,8 @@ async function runCommand(argv) {
 }
 async function main() {
   const argv = process.argv.slice(2);
-  if (argv.length === 0 || argv[0] !== "run" && argv[0] !== "live" && (argv.includes("--help") || argv.includes("-h"))) {
+  const knownCommands = ["run", "live", "watch"];
+  if (argv.length === 0 || !knownCommands.includes(argv[0]) && (argv.includes("--help") || argv.includes("-h"))) {
     printHelp();
     process.exit(0);
   }
@@ -1277,6 +1796,9 @@ async function main() {
     case "live":
       startLive(argv);
       break;
+    case "watch":
+      startWatch(argv);
+      break;
     default:
       if (!subcommand?.startsWith("-")) {
         startLive(["live", ...argv]);

package/dist/cli.js

@@ -1,8 +1,9 @@
 #!/usr/bin/env node
 import {
   runTraced,
-  startLive
-} from "./chunk-5SPZPOFN.js";
+  startLive,
+  startWatch
+} from "./chunk-NPH34CAL.js";
 
 // src/cli.ts
 import { basename } from "path";
@@ -14,8 +15,9 @@ Usage:
   agentflow <command> [options]
 
 Commands:
-  run   [options] -- <cmd>       Wrap a command with automatic execution tracing
-  live  [dir...] [options]      Real-time terminal monitor (auto-detects any JSON/JSONL)
+  run    [options] -- <cmd>       Wrap a command with automatic execution tracing
+  live   [dir...] [options]      Real-time terminal monitor (auto-detects any JSON/JSONL)
+  watch  [dir...] [options]      Headless alert system \u2014 detects failures, sends notifications
 
 Run \`agentflow <command> --help\` for command-specific options.
 
@@ -23,7 +25,8 @@ Examples:
   agentflow run --traces-dir ./traces -- python -m myagent process
   agentflow live ./data
   agentflow live ./traces ./cron ./workers -R
-  agentflow live ./data --refresh 5
+  agentflow watch ./data --alert-on error --notify telegram
+  agentflow watch ./data ./cron --alert-on stale:15m --notify webhook:https://...
 `.trim());
 }
 function parseRunArgs(argv) {
@@ -161,7 +164,8 @@ async function runCommand(argv) {
 }
 async function main() {
   const argv = process.argv.slice(2);
-  if (argv.length === 0 || argv[0] !== "run" && argv[0] !== "live" && (argv.includes("--help") || argv.includes("-h"))) {
+  const knownCommands = ["run", "live", "watch"];
+  if (argv.length === 0 || !knownCommands.includes(argv[0]) && (argv.includes("--help") || argv.includes("-h"))) {
     printHelp();
     process.exit(0);
   }
@@ -173,6 +177,9 @@ async function main() {
     case "live":
       startLive(argv);
       break;
+    case "watch":
+      startWatch(argv);
+      break;
     default:
       if (!subcommand?.startsWith("-")) {
         startLive(["live", ...argv]);