fix delegate timeout (#284)

* pin all CLI installations to explicit versions and use pro models by default

- codex: pin @openai/codex to 0.101.0 (was "latest")
- opencode: pin opencode-ai to 1.1.56 (was "latest")
- gemini: pin gemini-cli to v0.28.2 via new tag param on installFromGithub
- cursor: pin to 2026.01.28-fd13201 via direct tarball download (replaces curl install script)
- add installFromDirectTarball to install.ts for versioned tarball URLs
- gemini auto effort now uses pro-preview instead of flash-preview
- test runner model overrides updated to use pro-preview consistently

Co-authored-by: Cursor <cursoragent@cursor.com>

* increase activity timeout

* fix delegation timeout

* fix delegate timeouts

---------

Co-authored-by: Cursor <cursoragent@cursor.com>

agents/claude.ts

@@ -142,6 +142,7 @@ export const claude = agent({
       activityTimeout: 0, // disabled: process-level timeout in main.ts handles this (subprocess timeout would kill orchestrator during delegation)
       onStdout: async (chunk) => {
         finalOutput += chunk;
+        markActivity(); // reset activity timeout on any CLI output
 
         // buffer incomplete lines across chunks (NDJSON format)
         stdoutBuffer += chunk;

agents/codex.ts

@@ -232,6 +232,7 @@ export const codex = agent({
       activityTimeout: 0, // disabled: process-level timeout in main.ts handles this (subprocess timeout would kill orchestrator during delegation)
       onStdout: async (chunk) => {
         finalOutput += chunk;
+        markActivity(); // reset activity timeout on any CLI output
 
         // buffer incomplete lines across chunks (NDJSON format)
         stdoutBuffer += chunk;

agents/cursor.ts

@@ -274,6 +274,7 @@ export const cursor = agent({
 
         let stdout = "";
         let stderr = "";
+        let stdoutBuffer = "";
 
         child.on("spawn", () => {
           log.debug("Cursor CLI process spawned");
@@ -282,25 +283,36 @@ export const cursor = agent({
         child.stdout?.on("data", async (data) => {
           const text = data.toString();
           stdout += text;
+          markActivity(); // reset activity timeout on any CLI output
 
-          try {
+          // buffer incomplete lines across chunks (NDJSON format)
-            const event = JSON.parse(text) as CursorEvent;
+          stdoutBuffer += text;
-            markActivity(); // reset activity timeout on every event
+          const lines = stdoutBuffer.split("\n");
-            log.debug(JSON.stringify(event, null, 2));
 
-            // skip empty thinking deltas
+          // keep the last element (may be incomplete) in the buffer
-            if (event.type === "thinking" && event.subtype === "delta" && !event.text) {
+          stdoutBuffer = lines.pop() || "";
-              return;
+
+          for (const line of lines) {
+            const trimmed = line.trim();
+            if (!trimmed) continue;
+
+            try {
+              const event = JSON.parse(trimmed) as CursorEvent;
+              log.debug(JSON.stringify(event, null, 2));
+
+              // skip empty thinking deltas
+              if (event.type === "thinking" && event.subtype === "delta" && !event.text) {
+                continue;
+              }
+
+              // route to appropriate handler
+              const handler = messageHandlers[event.type as keyof typeof messageHandlers];
+              if (handler) {
+                await handler(event as never);
+              }
+            } catch {
+              // ignore parse errors - might be formatted tool call logs from cursor cli
             }
-
-            // route to appropriate handler
-            const handler = messageHandlers[event.type as keyof typeof messageHandlers];
-            if (handler) {
-              await handler(event as never);
-            }
-          } catch {
-            // ignore parse errors - might be formatted tool call logs from cursor cli
-            // our handlers log tool calls instead, so we don't need to display these
           }
         });
 

agents/gemini.ts

@@ -239,6 +239,7 @@ export const gemini = agent({
           onStdout: async (chunk) => {
             const text = chunk.toString();
             finalOutput += text;
+            markActivity(); // reset activity timeout on any CLI output
 
             // buffer incomplete lines across chunks (NDJSON format)
             stdoutBuffer += text;

agents/opencode.ts

@@ -123,6 +123,7 @@ export const opencode = agent({
         onStdout: async (chunk) => {
           const text = chunk.toString();
           output += text;
+          markActivity(); // reset activity timeout on any CLI output
 
           // buffer incomplete lines across chunks (NDJSON format)
           stdoutBuffer += text;

entry

@@ -141921,6 +141921,7 @@ function DelegateTool(ctx) {
       log.info(
         `\xBB delegating to ${selectedMode.name} mode (effort=${effort})${params.instructions ? " with orchestrator instructions" : ""}`
       );
+      const keepAliveInterval = setInterval(markActivity, 3e4);
       try {
         const subagentPayload = { ...ctx.payload, effort };
         const subagentInstructions = resolveSubagentInstructions({
@@ -141954,6 +141955,7 @@ function DelegateTool(ctx) {
           error: errorMessage
         };
       } finally {
+        clearInterval(keepAliveInterval);
         ctx.toolState.delegationActive = false;
       }
     })
@@ -144858,6 +144860,7 @@ var claude = agent({
       // disabled: process-level timeout in main.ts handles this (subprocess timeout would kill orchestrator during delegation)
       onStdout: async (chunk) => {
         finalOutput2 += chunk;
+        markActivity();
         stdoutBuffer += chunk;
         const lines = stdoutBuffer.split("\n");
         stdoutBuffer = lines.pop() || "";
@@ -145140,6 +145143,7 @@ var codex = agent({
       // disabled: process-level timeout in main.ts handles this (subprocess timeout would kill orchestrator during delegation)
       onStdout: async (chunk) => {
         finalOutput2 += chunk;
+        markActivity();
         stdoutBuffer += chunk;
         const lines = stdoutBuffer.split("\n");
         stdoutBuffer = lines.pop() || "";
@@ -145412,24 +145416,32 @@ var cursor = agent({
         });
         let stdout = "";
         let stderr = "";
+        let stdoutBuffer = "";
         child.on("spawn", () => {
           log.debug("Cursor CLI process spawned");
         });
         child.stdout?.on("data", async (data) => {
           const text = data.toString();
           stdout += text;
-          try {
+          markActivity();
-            const event = JSON.parse(text);
+          stdoutBuffer += text;
-            markActivity();
+          const lines = stdoutBuffer.split("\n");
-            log.debug(JSON.stringify(event, null, 2));
+          stdoutBuffer = lines.pop() || "";
-            if (event.type === "thinking" && event.subtype === "delta" && !event.text) {
+          for (const line of lines) {
-              return;
+            const trimmed = line.trim();
+            if (!trimmed) continue;
+            try {
+              const event = JSON.parse(trimmed);
+              log.debug(JSON.stringify(event, null, 2));
+              if (event.type === "thinking" && event.subtype === "delta" && !event.text) {
+                continue;
+              }
+              const handler2 = messageHandlers5[event.type];
+              if (handler2) {
+                await handler2(event);
+              }
+            } catch {
             }
-            const handler2 = messageHandlers5[event.type];
-            if (handler2) {
-              await handler2(event);
-            }
-          } catch {
           }
         });
         child.stderr?.on("data", (data) => {
@@ -145660,6 +145672,7 @@ var gemini = agent({
           onStdout: async (chunk) => {
             const text = chunk.toString();
             finalOutput2 += text;
+            markActivity();
             stdoutBuffer += text;
             const lines = stdoutBuffer.split("\n");
             stdoutBuffer = lines.pop() || "";
@@ -145863,6 +145876,7 @@ var opencode = agent({
         onStdout: async (chunk) => {
           const text = chunk.toString();
           output += text;
+          markActivity();
           stdoutBuffer += text;
           const lines = stdoutBuffer.split("\n");
           stdoutBuffer = lines.pop() || "";

mcp/delegate.ts

@@ -1,6 +1,7 @@
 import { type } from "arktype";
 import { Effort } from "../external.ts";
 import type { Mode } from "../modes.ts";
+import { markActivity } from "../utils/activity.ts";
 import { log } from "../utils/cli.ts";
 import { resolveSubagentInstructions } from "../utils/instructions.ts";
 import type { ToolContext } from "./server.ts";
@@ -71,6 +72,12 @@ export function DelegateTool(ctx: ToolContext) {
         `» delegating to ${selectedMode.name} mode (effort=${effort})${params.instructions ? " with orchestrator instructions" : ""}`
       );
 
+      // keep the process-level activity timeout alive while the subagent runs.
+      // agent CLIs can have long silent thinking phases (>60s) with no stdout,
+      // which would trigger the activity timeout. the overall run timeout (default 1h)
+      // is the real safety net for stalled agents.
+      const keepAliveInterval = setInterval(markActivity, 30_000);
+
       try {
         // build subagent payload with effort override
         const subagentPayload = { ...ctx.payload, effort };
@@ -112,6 +119,7 @@ export function DelegateTool(ctx: ToolContext) {
           error: errorMessage,
         };
       } finally {
+        clearInterval(keepAliveInterval);
         // always release the lock so the orchestrator can delegate again
         ctx.toolState.delegationActive = false;
       }

test/adhoc/delegateTimeout.ts

@@ -0,0 +1,56 @@
+import type { AgentResult, TestRunnerOptions, ValidationCheck } from "../utils.ts";
+import { defineFixture, getAgentOutput, getStructuredOutput } from "../utils.ts";
+
+/**
+ * delegateTimeout test - validates that the activity timeout does NOT fire
+ * during a delegation that takes longer than 60 seconds.
+ *
+ * uses effort: "auto" for both orchestrator and subagent so the total
+ * delegation time exceeds 60s. if the markActivity fix is missing,
+ * this test will fail with "activity timeout: no output for Xs".
+ */
+
+const fixture = defineFixture(
+  {
+    prompt: `Delegate to the Plan mode with auto effort. Pass these instructions to the subagent:
+"Carefully analyze the following engineering question and provide a thorough response, then call set_output with the value 'DELEGATE_TIMEOUT_PASSED'.
+
+Question: Design a comprehensive error handling strategy for a distributed microservices architecture. Consider:
+1. Circuit breaker patterns — when to open, half-open, close. What thresholds to use.
+2. Retry policies — exponential backoff with jitter. Maximum retry counts. Which errors are retryable.
+3. Dead letter queues — when to use them, how to process failed messages, alerting.
+4. Health check endpoints — liveness vs readiness probes, dependency health checks.
+5. Graceful degradation — fallback responses, feature flags, bulkhead pattern.
+
+Provide a detailed analysis covering ALL 5 points with concrete examples before calling set_output."`,
+    effort: "auto",
+    timeout: "8m",
+  },
+  { localOnly: true }
+);
+
+function validator(result: AgentResult): ValidationCheck[] {
+  const output = getStructuredOutput(result);
+  const agentOutput = getAgentOutput(result);
+
+  const setOutputCalled = output !== null;
+  const correctValue = setOutputCalled && /DELEGATE_TIMEOUT_PASSED/i.test(output);
+  const delegationOccurred = /» delegating to \w+ mode/i.test(agentOutput);
+  // the critical check: no activity timeout occurred
+  const noActivityTimeout = !/activity timeout/i.test(agentOutput);
+
+  return [
+    { name: "set_output", passed: setOutputCalled },
+    { name: "correct_value", passed: correctValue },
+    { name: "delegation_occurred", passed: delegationOccurred },
+    { name: "no_activity_timeout", passed: noActivityTimeout },
+  ];
+}
+
+export const test: TestRunnerOptions = {
+  name: "delegate-timeout",
+  fixture,
+  validator,
+  env: { GITHUB_REPOSITORY: "pullfrog/test-repo" },
+  tags: ["adhoc"],
+};