karajan-code 1.9.0 → 1.9.3

package/README.md

@@ -43,6 +43,7 @@ Instead of running one AI agent and manually reviewing its output, `kj` chains a
 - **Task decomposition** — triage detects when tasks should be split and recommends subtasks; with Planning Game integration, creates linked cards with sequential blocking
 - **Retry with backoff** — automatic recovery from transient API errors (429, 5xx) with exponential backoff and jitter
 - **Pipeline stage tracker** — cumulative progress view during `kj_run` showing which stages are done, running, or pending — both in CLI and via MCP events for real-time host rendering
+- **Planner observability guardrails** — continuous heartbeat/stall telemetry, configurable max-silence protection (`session.max_agent_silence_minutes`), and hard runtime cap (`session.max_planner_minutes`) to avoid long stuck planner runs
 - **Planning Game integration** — optionally pair with [Planning Game](https://github.com/AgenteIA-Geniova/planning-game) for agile project management (tasks, sprints, estimation) — like Jira, but open-source and XP-native
 
 > **Best with MCP** — Karajan Code is designed to be used as an MCP server inside your AI agent (Claude, Codex, etc.). The agent sends tasks to `kj_run`, gets real-time progress notifications, and receives structured results — no copy-pasting needed.
@@ -417,7 +418,7 @@ After `npm install -g karajan-code`, the MCP server is auto-registered in Claude
 | `kj_roles` | List roles or show role templates |
 | `kj_code` | Run coder-only mode |
 | `kj_review` | Run reviewer-only mode |
-| `kj_plan` | Generate implementation plan |
+| `kj_plan` | Generate implementation plan with heartbeat/stall telemetry and clearer diagnostics |
 
 ### Recommended Companion MCPs
 

package/docs/README.es.md

@@ -42,6 +42,7 @@ En lugar de ejecutar un agente de IA y revisar manualmente su output, `kj` encad
 - **Descomposicion de tareas** — triage detecta cuando una tarea debe dividirse y recomienda subtareas; con integracion Planning Game, crea cards vinculadas con bloqueo secuencial
 - **Retry con backoff** — recuperacion automatica ante errores transitorios de API (429, 5xx) con backoff exponencial y jitter
 - **Pipeline stage tracker** — vista de progreso acumulativo durante `kj_run` mostrando que stages estan completadas, en ejecucion o pendientes — tanto en CLI como via eventos MCP para renderizado en tiempo real en el host
+- **Guardarrailes de observabilidad del planner** — telemetria continua de heartbeat/stall, proteccion configurable por silencio maximo (`session.max_agent_silence_minutes`) y limite duro de ejecucion (`session.max_planner_minutes`) para evitar bloqueos prolongados en `kj_plan`/planner
 - **Integracion con Planning Game** — combina opcionalmente con [Planning Game](https://github.com/AgenteIA-Geniova/planning-game) para gestion agil de proyectos (tareas, sprints, estimacion) — como Jira, pero open-source y nativo XP
 
 > **Mejor con MCP** — Karajan Code esta disenado para usarse como servidor MCP dentro de tu agente de IA (Claude, Codex, etc.). El agente envia tareas a `kj_run`, recibe notificaciones de progreso en tiempo real, y obtiene resultados estructurados — sin copiar y pegar.
@@ -201,7 +202,7 @@ Tras `npm install -g karajan-code`, el servidor MCP se auto-registra en las conf
 | `kj_roles` | Listar roles o mostrar templates |
 | `kj_code` | Modo solo coder |
 | `kj_review` | Modo solo reviewer |
-| `kj_plan` | Generar plan de implementacion |
+| `kj_plan` | Generar plan de implementacion con telemetria heartbeat/stall y diagnostico mas claro |
 
 ### MCPs complementarios recomendados
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "karajan-code",
-  "version": "1.9.0",
+  "version": "1.9.3",
   "description": "Local multi-agent coding orchestrator with TDD, SonarQube, and code review pipeline",
   "type": "module",
   "license": "AGPL-3.0",

package/src/agents/aider-agent.js

@@ -8,7 +8,11 @@ export class AiderAgent extends BaseAgent {
     const args = ["--yes", "--message", task.prompt];
     const model = this.getRoleModel(role);
     if (model) args.push("--model", model);
-    const res = await runCommand(resolveBin("aider"), args, { onOutput: task.onOutput });
+    const res = await runCommand(resolveBin("aider"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 
@@ -17,7 +21,11 @@ export class AiderAgent extends BaseAgent {
     const args = ["--yes", "--message", task.prompt];
     const model = this.getRoleModel(role);
     if (model) args.push("--model", model);
-    const res = await runCommand(resolveBin("aider"), args, { onOutput: task.onOutput });
+    const res = await runCommand(resolveBin("aider"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 }

package/src/agents/claude-agent.js

@@ -81,7 +81,11 @@ export class ClaudeAgent extends BaseAgent {
     if (task.onOutput) {
       args.push("--output-format", "stream-json");
       const streamFilter = createStreamJsonFilter(task.onOutput);
-      const res = await runCommand(resolveBin("claude"), args, { onOutput: streamFilter });
+      const res = await runCommand(resolveBin("claude"), args, {
+        onOutput: streamFilter,
+        silenceTimeoutMs: task.silenceTimeoutMs,
+        timeout: task.timeoutMs
+      });
       const output = extractTextFromStreamJson(res.stdout);
       return { ok: res.exitCode === 0, output, error: res.stderr, exitCode: res.exitCode };
     }
@@ -94,7 +98,11 @@ export class ClaudeAgent extends BaseAgent {
     const args = ["-p", task.prompt, "--output-format", "json"];
     const model = this.getRoleModel(task.role || "reviewer");
     if (model) args.push("--model", model);
-    const res = await runCommand(resolveBin("claude"), args, { onOutput: task.onOutput });
+    const res = await runCommand(resolveBin("claude"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 }

package/src/agents/codex-agent.js

@@ -9,8 +9,13 @@ export class CodexAgent extends BaseAgent {
     const model = this.getRoleModel(role);
     if (model) args.push("--model", model);
     if (this.isAutoApproveEnabled(role)) args.push("--full-auto");
-    args.push(task.prompt);
-    const res = await runCommand(resolveBin("codex"), args, { onOutput: task.onOutput });
+    args.push("-");
+    const res = await runCommand(resolveBin("codex"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs,
+      input: task.prompt
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 
@@ -18,8 +23,13 @@ export class CodexAgent extends BaseAgent {
     const args = ["exec"];
     const model = this.getRoleModel(task.role || "reviewer");
     if (model) args.push("--model", model);
-    args.push(task.prompt);
-    const res = await runCommand(resolveBin("codex"), args, { onOutput: task.onOutput });
+    args.push("-");
+    const res = await runCommand(resolveBin("codex"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs,
+      input: task.prompt
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 }

package/src/agents/gemini-agent.js

@@ -8,7 +8,11 @@ export class GeminiAgent extends BaseAgent {
     const args = ["-p", task.prompt];
     const model = this.getRoleModel(role);
     if (model) args.push("--model", model);
-    const res = await runCommand(resolveBin("gemini"), args, { onOutput: task.onOutput });
+    const res = await runCommand(resolveBin("gemini"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 
@@ -17,7 +21,11 @@ export class GeminiAgent extends BaseAgent {
     const args = ["-p", task.prompt, "--output-format", "json"];
     const model = this.getRoleModel(role);
     if (model) args.push("--model", model);
-    const res = await runCommand(resolveBin("gemini"), args, { onOutput: task.onOutput });
+    const res = await runCommand(resolveBin("gemini"), args, {
+      onOutput: task.onOutput,
+      silenceTimeoutMs: task.silenceTimeoutMs,
+      timeout: task.timeoutMs
+    });
     return { ok: res.exitCode === 0, output: res.stdout, error: res.stderr, exitCode: res.exitCode };
   }
 }

package/src/cli.js

@@ -187,6 +187,29 @@ program
     });
   });
 
+program
+  .command("update")
+  .description("Update karajan-code to the latest version from npm")
+  .action(async () => {
+    const { execaCommand } = await import("execa");
+    console.log(`Current version: ${PKG_VERSION}`);
+    console.log("Checking for updates...");
+    try {
+      const { stdout } = await execaCommand("npm view karajan-code version");
+      const latest = stdout.trim();
+      if (latest === PKG_VERSION) {
+        console.log(`Already on the latest version (${PKG_VERSION}).`);
+        return;
+      }
+      console.log(`Updating ${PKG_VERSION} → ${latest}...`);
+      await execaCommand("npm install -g karajan-code@latest", { stdio: "inherit" });
+      console.log(`Updated to ${latest}. Restart Claude to pick up the new MCP server.`);
+    } catch (err) {
+      console.error(`Update failed: ${err.message}`);
+      process.exit(1);
+    }
+  });
+
 const sonar = program.command("sonar").description("Manage SonarQube container");
 sonar.command("status").action(async () => sonarCommand({ action: "status" }));
 sonar.command("start").action(async () => sonarCommand({ action: "start" }));

package/src/commands/plan.js

@@ -46,7 +46,13 @@ export async function planCommand({ task, config, logger, json, context }) {
 
   const planner = createAgent(plannerRole.provider, config, logger);
   const prompt = buildPlannerPrompt({ task, context });
-  const result = await planner.runTask({ prompt, role: "planner" });
+  const silenceTimeoutMs = Number(config?.session?.max_agent_silence_minutes) > 0
+    ? Math.round(Number(config.session.max_agent_silence_minutes) * 60 * 1000)
+    : undefined;
+  const timeoutMs = Number(config?.session?.max_planner_minutes) > 0
+    ? Math.round(Number(config.session.max_planner_minutes) * 60 * 1000)
+    : undefined;
+  const result = await planner.runTask({ prompt, role: "planner", silenceTimeoutMs, timeoutMs });
 
   if (!result.ok) {
     throw new Error(result.error || result.output || "Planner failed");

package/src/config.js

@@ -114,7 +114,9 @@ const DEFAULTS = {
   session: {
     max_iteration_minutes: 30,
     max_total_minutes: 120,
+    max_planner_minutes: 60,
     checkpoint_interval_minutes: 5,
+    max_agent_silence_minutes: 20,
     fail_fast_repeats: 2,
     repeat_detection_threshold: 2,
     max_sonar_retries: 3,

package/src/mcp/server-handlers.js

@@ -62,6 +62,18 @@ export function classifyError(error) {
   const msg = error?.message || String(error);
   const lower = msg.toLowerCase();
 
+  if (
+    lower.includes("without output")
+    || lower.includes("silent for")
+    || lower.includes("unresponsive")
+    || lower.includes("exceeded max silence")
+  ) {
+    return {
+      category: "agent_stall",
+      suggestion: "Agent output stalled. Check live details with kj_status, then retry with a smaller prompt or increase session.max_agent_silence_minutes if needed."
+    };
+  }
+
   if (lower.includes("sonar") && (lower.includes("connect") || lower.includes("econnrefused") || lower.includes("not available") || lower.includes("not running"))) {
     return {
       category: "sonar_unavailable",
@@ -235,7 +247,15 @@ export async function handlePlanDirect(a, server, extra) {
 
   const projectDir = await resolveProjectDir(server);
   const runLog = createRunLog(projectDir);
-  runLog.logText(`[kj_plan] started — provider=${plannerRole.provider}`);
+  const silenceTimeoutMs = Number(config?.session?.max_agent_silence_minutes) > 0
+    ? Math.round(Number(config.session.max_agent_silence_minutes) * 60 * 1000)
+    : undefined;
+  const plannerTimeoutMs = Number(config?.session?.max_planner_minutes) > 0
+    ? Math.round(Number(config.session.max_planner_minutes) * 60 * 1000)
+    : undefined;
+  runLog.logText(
+    `[kj_plan] started — provider=${plannerRole.provider}, max_silence=${silenceTimeoutMs ? `${Math.round(silenceTimeoutMs / 1000)}s` : "disabled"}, max_runtime=${plannerTimeoutMs ? `${Math.round(plannerTimeoutMs / 1000)}s` : "disabled"}`
+  );
   const emitter = buildDirectEmitter(server, runLog);
   const eventBase = { sessionId: null, iteration: 0, startedAt: Date.now() };
   const onOutput = ({ stream, line }) => {
@@ -250,18 +270,31 @@ export async function handlePlanDirect(a, server, extra) {
   sendTrackerLog(server, "planner", "running", plannerRole.provider);
   runLog.logText(`[planner] agent launched, waiting for response...`);
   let result;
+  let plannerStats = null;
   try {
-    result = await planner.runTask({ prompt, role: "planner", onOutput: stallDetector.onOutput });
+    result = await planner.runTask({
+      prompt,
+      role: "planner",
+      onOutput: stallDetector.onOutput,
+      silenceTimeoutMs,
+      timeoutMs: plannerTimeoutMs
+    });
   } finally {
     stallDetector.stop();
-    const stats = stallDetector.stats();
-    runLog.logText(`[planner] finished — lines=${stats.lineCount}, bytes=${stats.bytesReceived}, elapsed=${Math.round(stats.elapsedMs / 1000)}s`);
+    plannerStats = stallDetector.stats();
+    runLog.logText(
+      `[planner] finished — lines=${plannerStats.lineCount}, bytes=${plannerStats.bytesReceived}, elapsed=${Math.round(plannerStats.elapsedMs / 1000)}s`
+    );
     runLog.close();
   }
 
   if (!result.ok) {
     sendTrackerLog(server, "planner", "failed");
-    throw new Error(result.error || result.output || "Planner failed");
+    const baseError = result.error || result.output || "Planner failed";
+    const statsSuffix = plannerStats
+      ? ` [lines=${plannerStats.lineCount}, bytes=${plannerStats.bytesReceived}, elapsed=${Math.round(plannerStats.elapsedMs / 1000)}s]`
+      : "";
+    throw new Error(`${baseError}${statsSuffix}`);
   }
 
   sendTrackerLog(server, "planner", "done");

package/src/roles/planner-role.js

@@ -9,6 +9,18 @@ function resolveProvider(config) {
   );
 }
 
+function resolvePlannerSilenceTimeoutMs(config) {
+  const minutes = Number(config?.session?.max_agent_silence_minutes);
+  if (!Number.isFinite(minutes) || minutes <= 0) return null;
+  return Math.round(minutes * 60 * 1000);
+}
+
+function resolvePlannerRuntimeTimeoutMs(config) {
+  const minutes = Number(config?.session?.max_planner_minutes);
+  if (!Number.isFinite(minutes) || minutes <= 0) return null;
+  return Math.round(minutes * 60 * 1000);
+}
+
 function buildPrompt({ task, instructions, research, triageDecomposition }) {
   const sections = [];
 
@@ -78,6 +90,10 @@ export class PlannerRole extends BaseRole {
 
     const runArgs = { prompt, role: "planner" };
     if (onOutput) runArgs.onOutput = onOutput;
+    const silenceTimeoutMs = resolvePlannerSilenceTimeoutMs(this.config);
+    if (silenceTimeoutMs) runArgs.silenceTimeoutMs = silenceTimeoutMs;
+    const timeoutMs = resolvePlannerRuntimeTimeoutMs(this.config);
+    if (timeoutMs) runArgs.timeoutMs = timeoutMs;
     const result = await agent.runTask(runArgs);
 
     if (!result.ok) {

package/src/utils/process.js

@@ -1,7 +1,7 @@
 import { execa } from "execa";
 
 export async function runCommand(command, args = [], options = {}) {
-  const { timeout, onOutput, ...rest } = options;
+  const { timeout, onOutput, silenceTimeoutMs, partialOutputFlushMs, ...rest } = options;
   const subprocess = execa(command, args, {
     reject: false,
     ...rest
@@ -9,37 +9,98 @@ export async function runCommand(command, args = [], options = {}) {
 
   let stdoutAccum = "";
   let stderrAccum = "";
+  let outputSilenceTimer = null;
+  let silenceTimedOut = false;
+
+  function clearSilenceTimer() {
+    if (outputSilenceTimer) {
+      clearTimeout(outputSilenceTimer);
+      outputSilenceTimer = null;
+    }
+  }
+
+  function armSilenceTimer() {
+    const ms = Number(silenceTimeoutMs);
+    if (!Number.isFinite(ms) || ms <= 0 || silenceTimedOut) return;
+    clearSilenceTimer();
+    outputSilenceTimer = setTimeout(() => {
+      silenceTimedOut = true;
+      try {
+        subprocess.kill("SIGKILL", { forceKillAfterDelay: 1000 });
+      } catch {
+        // no-op
+      }
+    }, ms);
+  }
 
   if (subprocess.stdout) {
     subprocess.stdout.on("data", (chunk) => {
       stdoutAccum += chunk.toString();
+      armSilenceTimer();
     });
   }
   if (subprocess.stderr) {
     subprocess.stderr.on("data", (chunk) => {
       stderrAccum += chunk.toString();
+      armSilenceTimer();
     });
   }
 
+  let flushInterval = null;
   if (onOutput) {
-    const handler = (stream) => {
-      let partial = "";
+    const flushMs = Number(partialOutputFlushMs) > 0 ? Number(partialOutputFlushMs) : 2000;
+    const streams = {};
+    const makeHandler = (stream) => {
+      const state = { partial: "", dirty: false };
+      streams[stream] = state;
       return (chunk) => {
-        partial += chunk.toString();
-        const lines = partial.split("\n");
-        partial = lines.pop();
+        state.partial += chunk.toString();
+        const lines = state.partial.split(/\r\n|\n|\r/);
+        state.partial = lines.pop() ?? "";
+        state.dirty = state.partial.length > 0;
         for (const line of lines) {
           if (line) onOutput({ stream, line });
         }
       };
     };
-    if (subprocess.stdout) subprocess.stdout.on("data", handler("stdout"));
-    if (subprocess.stderr) subprocess.stderr.on("data", handler("stderr"));
+
+    const flushPartials = () => {
+      for (const [stream, state] of Object.entries(streams)) {
+        if (!state.dirty || !state.partial) continue;
+        onOutput({ stream, line: state.partial });
+        state.partial = "";
+        state.dirty = false;
+      }
+    };
+
+    if (subprocess.stdout) subprocess.stdout.on("data", makeHandler("stdout"));
+    if (subprocess.stderr) subprocess.stderr.on("data", makeHandler("stderr"));
+    flushInterval = setInterval(flushPartials, flushMs);
+    flushInterval.unref?.();
+
+    subprocess.finally(() => {
+      flushPartials();
+      if (flushInterval) {
+        clearInterval(flushInterval);
+        flushInterval = null;
+      }
+    });
   }
+  armSilenceTimer();
 
   try {
     if (!timeout) {
       const result = await subprocess;
+      clearSilenceTimer();
+      if (silenceTimedOut) {
+        return {
+          exitCode: 143,
+          stdout: stdoutAccum,
+          stderr: `Command killed after ${Number(silenceTimeoutMs)}ms without output`,
+          timedOut: true,
+          signal: "SIGKILL"
+        };
+      }
       return enrichResult(result, stdoutAccum, stderrAccum);
     }
 
@@ -63,8 +124,28 @@ export async function runCommand(command, args = [], options = {}) {
 
     const result = await Promise.race([subprocess, timeoutResult]);
     if (timer) clearTimeout(timer);
+    clearSilenceTimer();
+    if (silenceTimedOut) {
+      return {
+        exitCode: 143,
+        stdout: stdoutAccum,
+        stderr: `Command killed after ${Number(silenceTimeoutMs)}ms without output`,
+        timedOut: true,
+        signal: "SIGKILL"
+      };
+    }
     return enrichResult(result, stdoutAccum, stderrAccum);
   } catch (error) {
+    clearSilenceTimer();
+    if (silenceTimedOut) {
+      return {
+        exitCode: 143,
+        stdout: error?.stdout || stdoutAccum,
+        stderr: `Command killed after ${Number(silenceTimeoutMs)}ms without output`,
+        timedOut: true,
+        signal: error?.signal || "SIGKILL"
+      };
+    }
     const details = [
       error?.shortMessage,
       error?.originalMessage,

package/src/utils/stall-detector.js

@@ -15,6 +15,7 @@ import { emitProgress, makeEvent } from "./events.js";
 const DEFAULT_HEARTBEAT_INTERVAL_MS = 30_000;   // heartbeat every 30s
 const DEFAULT_STALL_TIMEOUT_MS      = 120_000;  // warn after 2min silence
 const DEFAULT_CRITICAL_TIMEOUT_MS   = 300_000;  // critical after 5min silence
+const DEFAULT_STALL_REPEAT_MS       = 60_000;   // repeat stall notices every 60s
 
 export function createStallDetector({
   onOutput,
@@ -24,23 +25,30 @@ export function createStallDetector({
   provider,
   heartbeatIntervalMs = DEFAULT_HEARTBEAT_INTERVAL_MS,
   stallTimeoutMs      = DEFAULT_STALL_TIMEOUT_MS,
-  criticalTimeoutMs   = DEFAULT_CRITICAL_TIMEOUT_MS
+  criticalTimeoutMs   = DEFAULT_CRITICAL_TIMEOUT_MS,
+  stallRepeatMs       = DEFAULT_STALL_REPEAT_MS,
+  maxSilenceMs        = null,
+  onMaxSilence        = null
 }) {
   let lastActivityAt = Date.now();
   let lineCount = 0;
   let bytesReceived = 0;
-  let stallWarned = false;
-  let criticalWarned = false;
   let heartbeatTimer = null;
   const startedAt = Date.now();
+  let lastStallWarnAt = 0;
+  let lastCriticalWarnAt = 0;
+  let maxSilenceTriggered = false;
 
   function emitHeartbeat() {
     const now = Date.now();
     const silenceMs = now - lastActivityAt;
     const elapsedMs = now - startedAt;
+    const shouldWarn = silenceMs >= stallTimeoutMs;
+    const shouldCritical = silenceMs >= criticalTimeoutMs;
+    const repeatWindow = Math.max(1000, Number(stallRepeatMs) || DEFAULT_STALL_REPEAT_MS);
 
-    if (silenceMs >= criticalTimeoutMs && !criticalWarned) {
-      criticalWarned = true;
+    if (shouldCritical && (now - lastCriticalWarnAt >= repeatWindow)) {
+      lastCriticalWarnAt = now;
       emitProgress(emitter, makeEvent("agent:stall", { ...eventBase, stage }, {
         status: "critical",
         message: `Agent ${provider} unresponsive for ${Math.round(silenceMs / 1000)}s — may be hung`,
@@ -53,8 +61,8 @@ export function createStallDetector({
           severity: "critical"
         }
       }));
-    } else if (silenceMs >= stallTimeoutMs && !stallWarned) {
-      stallWarned = true;
+    } else if (shouldWarn && (now - lastStallWarnAt >= repeatWindow)) {
+      lastStallWarnAt = now;
       emitProgress(emitter, makeEvent("agent:stall", { ...eventBase, stage }, {
         status: "warning",
         message: `Agent ${provider} silent for ${Math.round(silenceMs / 1000)}s — still waiting`,
@@ -67,20 +75,49 @@ export function createStallDetector({
           severity: "warning"
         }
       }));
-    } else if (silenceMs < stallTimeoutMs) {
-      // Reset warning flags when activity resumes
-      stallWarned = false;
-      criticalWarned = false;
+    }
+
+    emitProgress(emitter, makeEvent("agent:heartbeat", { ...eventBase, stage }, {
+      message: silenceMs < stallTimeoutMs
+        ? `Agent ${provider} active — ${lineCount} lines, ${Math.round(elapsedMs / 1000)}s elapsed`
+        : `Agent ${provider} waiting — silent ${Math.round(silenceMs / 1000)}s, ${Math.round(elapsedMs / 1000)}s elapsed`,
+      detail: {
+        provider,
+        elapsedMs,
+        silenceMs,
+        lineCount,
+        bytesReceived,
+        status: silenceMs < stallTimeoutMs ? "active" : "waiting"
+      }
+    }));
 
-      emitProgress(emitter, makeEvent("agent:heartbeat", { ...eventBase, stage }, {
-        message: `Agent ${provider} active — ${lineCount} lines, ${Math.round(elapsedMs / 1000)}s elapsed`,
+    const hardLimit = Number(maxSilenceMs);
+    if (!maxSilenceTriggered && Number.isFinite(hardLimit) && hardLimit > 0 && silenceMs >= hardLimit) {
+      maxSilenceTriggered = true;
+      emitProgress(emitter, makeEvent("agent:stall", { ...eventBase, stage }, {
+        status: "fail",
+        message: `Agent ${provider} exceeded max silence (${Math.round(hardLimit / 1000)}s)`,
         detail: {
           provider,
+          silenceMs,
           elapsedMs,
           lineCount,
-          bytesReceived
+          bytesReceived,
+          severity: "fatal",
+          maxSilenceMs: hardLimit
         }
       }));
+      if (typeof onMaxSilence === "function") {
+        onMaxSilence({
+          provider,
+          stage,
+          silenceMs,
+          elapsedMs,
+          lineCount,
+          bytesReceived,
+          maxSilenceMs: hardLimit
+        });
+      }
     }
   }
 
@@ -92,10 +129,6 @@ export function createStallDetector({
     lineCount++;
     bytesReceived += data.line?.length || 0;
 
-    // Reset stall flags on new activity
-    stallWarned = false;
-    criticalWarned = false;
-
     // Forward to the original callback
     if (onOutput) {
       onOutput(data);