From 3f30ef5b47e2c1e9b5e5324bd62973853cb57cd1 Mon Sep 17 00:00:00 2001
From: Matthew Wright <mattwright1186@gmail.com>
Date: Sat, 4 Jul 2026 17:28:30 -0500
Subject: [PATCH] persist sessionId on failure so a failed run stays resumable

runTaskAgent persisted the engine sessionId only after the success/failure branch, so a failed run lost its session and a retry had to rebuild context from scratch. Persist it before the branch, and return telemetry on the failure path too.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../agent-3-task/src/run-task-agent.spec.ts     |  9 +++++++++
 .../packages/agent-3-task/src/run-task-agent.ts | 17 ++++++++++-------
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/orchestrator-v2/packages/agent-3-task/src/run-task-agent.spec.ts b/orchestrator-v2/packages/agent-3-task/src/run-task-agent.spec.ts
index d6c4725..a166584 100644
--- a/orchestrator-v2/packages/agent-3-task/src/run-task-agent.spec.ts
+++ b/orchestrator-v2/packages/agent-3-task/src/run-task-agent.spec.ts
@@ -132,6 +132,15 @@ describe("runTaskAgent", () => {
     then: { engine_failure_returned },
   });
 
+  test("persists sessionId and returns telemetry when the engine fails (resumable retry)", {
+    given: { failing_engine, valid_state_context },
+    when: { task_agent_run },
+    then: {
+      session_id_is_persisted,
+      telemetry_is_returned,
+    },
+  });
+
   test("fails when engine throws", {
     given: { throwing_engine, valid_state_context },
     when: { task_agent_run },
diff --git a/orchestrator-v2/packages/agent-3-task/src/run-task-agent.ts b/orchestrator-v2/packages/agent-3-task/src/run-task-agent.ts
index 58e5de3..41e0e25 100644
--- a/orchestrator-v2/packages/agent-3-task/src/run-task-agent.ts
+++ b/orchestrator-v2/packages/agent-3-task/src/run-task-agent.ts
@@ -48,13 +48,8 @@ export async function runTaskAgent(
     return { outcome: "failed", message };
   }
 
-  if (!engineResult.success) {
-    return {
-      outcome: "failed",
-      message: engineResult.lastMessage ?? "Engine execution failed",
-    };
-  }
-
+  // Persist the sessionId before branching on the outcome, so a failed run stays
+  // resumable — a retry can --resume instead of rebuilding context from scratch.
   if (engineResult.sessionId !== undefined) {
     await ctx.setState({
       ...workItem.state,
@@ -62,6 +57,14 @@ export async function runTaskAgent(
     });
   }
 
+  if (!engineResult.success) {
+    return {
+      outcome: "failed",
+      message: engineResult.lastMessage ?? "Engine execution failed",
+      telemetry: engineResult.stats ?? undefined,
+    };
+  }
+
   return {
     outcome: "completed",
     message: engineResult.lastMessage ?? undefined,