fix(orchestration): say what died in the worker-exit escalation

The escalation carried a subject line and a raw payload and nothing else, so the
banner a coordinator actually reads was a code and three ids. Learning which task
had died, or whether it was still retryable, meant looking the ids up by hand.

Add a body naming the task and stating the resulting state, taken from what
failDispatch settled to: a plain failure says the task is ready to dispatch again,
a tripped circuit breaker says it will not be retried automatically. Task naming
reuses buildOrchestrationTaskDisplayMetadata, so it matches the title shown
everywhere else rather than inventing a second derivation.

Before:
  Subject: Agent exited unexpectedly (code 137)
  [Payload: {"taskId":"task_c917b402bd49",...}]

After:
  Subject: Agent exited unexpectedly (code 137)
  Worker term_62dc5050 exited with code 137 while running task
  "Migrate the billing schema to v2" (task_c917b402bd49). The task is ready to be
  dispatched again.
  [Payload: {"taskId":"task_c917b402bd49",...}]

Both bodies are pinned by tests; the grading oracle now compares the body so the
wording cannot rot silently.

src/main/runtime/orca-runtime.ts

@@ -128,6 +128,7 @@ import { mkdir, readFile, readdir, rm, stat } from 'node:fs/promises'
 import { resolveWorktreeCreateBase } from '../worktree-create-base'
 import { resolveWorktreeAddBaseRef } from '../../shared/worktree/base-ref'
 import { OrchestrationDb } from './orchestration/db'
+import type { DispatchStatus } from './orchestration/types'
 import { reconcileRequestedWorkerTerminalReleases } from './orchestration/worker-terminal-release-reconciliation'
 import {
   classifyWorkerTerminalProcessIncarnation,
@@ -16830,7 +16831,9 @@ export class OrcaRuntimeService {
     }
 
     const errorContext = `Agent exited with code ${exitCode}`
-    this._orchestrationDb.failDispatch(dispatch.id, errorContext, { workerProcessExited: true })
+    const settled = this._orchestrationDb.failDispatch(dispatch.id, errorContext, {
+      workerProcessExited: true
+    })
 
     // Why: create an escalation message so the coordinator is notified about
     // the unexpected exit, even if the circuit breaker hasn't tripped yet.
@@ -16847,6 +16850,7 @@ export class OrcaRuntimeService {
         from: handle,
         to: recipient.to,
         subject: `Agent exited unexpectedly (code ${exitCode})`,
+        body: this.describeWorkerExit(dispatch, exitCode, handle, settled?.status),
         type: 'escalation',
         priority: 'high',
         // Why: applyEscalationToDispatch rejects an escalation without an exact Dispatch
@@ -16872,6 +16876,33 @@ export class OrcaRuntimeService {
     }
   }
 
+  // Why: the banner shows the subject and a raw payload, so without prose the coordinator
+  // has to resolve ids by hand to learn what died and whether the task is still retryable.
+  private describeWorkerExit(
+    dispatch: { id: string; task_id: string; run_id: string },
+    exitCode: number,
+    handle: string,
+    settledStatus: DispatchStatus | undefined
+  ): string {
+    const task = this._orchestrationDb?.getTask?.(dispatch.task_id, dispatch.run_id)
+    const title =
+      typeof task?.spec === 'string'
+        ? buildOrchestrationTaskDisplayMetadata({
+            spec: task.spec,
+            taskTitle: task.task_title,
+            displayName: task.display_name
+          }).taskTitle
+        : ''
+    const named = title ? `"${title}" (${dispatch.task_id})` : dispatch.task_id
+    const outcome =
+      settledStatus === 'circuit_broken'
+        ? ' This task has now failed too many times, so it will not be retried automatically.'
+        : settledStatus === 'failed'
+          ? ' The task is ready to be dispatched again.'
+          : ''
+    return `Worker ${handle} exited with code ${exitCode} while running task ${named}.${outcome}`
+  }
+
   // Why: a lightweight Run keeps its coordinator in runs/run_coordinator_handles and
   // never writes the legacy coordinator_runs table, so gating solely on that table
   // dropped every worker-death escalation (STA-4604). Address the Run mailbox the

src/main/runtime/orchestration/lightweight-run-worker-exit-escalation.test.ts

@@ -113,6 +113,7 @@ async function gradeWorkerExit(
   escalation: Record<string, unknown> | null
   workerHandle: string
   runId: string
+  taskId: string
 }> {
   const { runtime, workerHandle, coordinatorHandle } = makeRuntimeWithTwoPanes()
   const db = new OrchestrationDb(':memory:')
@@ -154,12 +155,14 @@ async function gradeWorkerExit(
             type: escalation.type,
             priority: escalation.priority,
             subject: escalation.subject,
+            body: escalation.body,
             deliveryContract: escalation.delivery_contract,
             payload: JSON.parse(escalation.payload ?? 'null')
           }
         : null,
       workerHandle,
-      runId
+      runId,
+      taskId: task.id
     }
   } finally {
     db.close()
@@ -183,6 +186,9 @@ describe('STA-4604 worker PTY exit escalation reaches the coordinator', () => {
       type: 'escalation',
       priority: 'high',
       subject: 'Agent exited unexpectedly (code 137)',
+      body:
+        `Worker ${graded.workerHandle} exited with code 137 while running task ` +
+        `"do the work" (${graded.taskId}). The task is ready to be dispatched again.`,
       deliveryContract: 'current_delivery',
       payload: {
         taskId: expect.any(String),
@@ -349,10 +355,18 @@ describe('STA-4604 worker PTY exit escalation reaches the coordinator', () => {
       runtime.onPtyExit(WORKER_PTY_ID, 137)
       await settle()
 
+      const escalations = db.getUnreadRunMailbox(run.id, 100, ['escalation'])
       expect({
         dispatch: db.getDispatchContextById(dispatch.id)?.status,
-        escalations: db.getUnreadRunMailbox(run.id, 100, ['escalation']).length
+        escalations: escalations.length
       }).toEqual({ dispatch: 'circuit_broken', escalations: 1 })
+      // Why: a tripped breaker means nobody will retry it, so the body must not
+      // tell the coordinator the task is ready to go again.
+      expect(escalations.at(-1)?.body).toBe(
+        `Worker ${workerHandle} exited with code 137 while running task ` +
+          `"repeatedly failing work" (${task.id}). This task has now failed too many ` +
+          `times, so it will not be retried automatically.`
+      )
     } finally {
       db.close()
     }