workspace-server/internal/handlers/a2a_proxy.go

@@ -132,15 +132,20 @@ func readBodyWithLimit(r io.Reader, limit int, kind string) ([]byte, error) {
 //     a generic 502 page to canvas. 10s is well above realistic intra-region
 //     latencies and well below CF's edge timeout.
 //
-//  3. Transport.ResponseHeaderTimeout — 5min default. From request-body-end
+//  3. Transport.ResponseHeaderTimeout — 30min default. From request-body-end
 //     to response-headers-start. Configurable via
 //     A2A_PROXY_RESPONSE_HEADER_TIMEOUT (envx.Duration). Covers cold-start
-//     first-byte (30-60s OAuth flow above) with enough room for Opus agent
-//     turns and Codex scheduled tasks (big context + internal delegate_task
-//     round-trips routinely exceed the old 60s/180s ceilings). Body streaming
-//     after headers is governed by the
-//     per-request context deadline, NOT this timeout — so multi-minute agent
-//     responses still work fine.
+//     first-byte (30-60s OAuth flow above) AND long SYNCHRONOUS autonomous
+//     turns where the runtime computes the whole response before sending
+//     headers (it does NOT always stream a 200 early). A real "migrate from
+//     blob" turn ran 443s and was killed at the old 5min default with
+//     `timeout awaiting response headers` (core#2723 class — the SAME long-turn
+//     cut as the idle watchdog). Aligned to 30min = the agent-to-agent
+//     ceiling + the canvas idle default, so no LEGIT turn (none exceed that)
+//     trips it; a genuinely-stuck agent is surfaced by DialContext (fast,
+//     connection-level) + the reactive-health/heartbeat path, not by cutting
+//     a working turn short. Body streaming after headers is governed by the
+//     per-request context deadline, NOT this timeout.
 //
 // The point of (2) and (3) is to surface a *structured* 503 from
 // handleA2ADispatchError when the workspace agent is unreachable, so canvas
@@ -153,7 +158,7 @@ var a2aClient = &http.Client{
 			Timeout:   10 * time.Second,
 			KeepAlive: 30 * time.Second,
 		}).DialContext,
-		ResponseHeaderTimeout: envx.Duration("A2A_PROXY_RESPONSE_HEADER_TIMEOUT", 5*time.Minute),
+		ResponseHeaderTimeout: envx.Duration("A2A_PROXY_RESPONSE_HEADER_TIMEOUT", 30*time.Minute),
 		TLSHandshakeTimeout:   10 * time.Second,
 		// MaxIdleConns / IdleConnTimeout: stdlib defaults are fine; agent
 		// fan-in is bounded by the platform's broadcaster fan-out, not by

workspace-server/internal/handlers/a2a_proxy_test.go

@@ -2850,7 +2850,10 @@ func TestLookupDeliveryMode_ContextCanceled_FailsClosed(t *testing.T) {
 // ==================== a2aClient ResponseHeaderTimeout config ====================
 
 func TestA2AClientResponseHeaderTimeout(t *testing.T) {
-	const defaultTimeout = 5 * time.Minute
+	// core#2723 class: raised 5min→30min so long synchronous autonomous turns
+	// (e.g. a 443s "migrate from blob" run that errored with "timeout awaiting
+	// response headers") aren't cut short. Aligned to the agent ceiling + idle.
+	const defaultTimeout = 30 * time.Minute
 
 	// Default (unset env) — a2aClient was initialised at package load time.
 	if a2aClient.Transport.(*http.Transport).ResponseHeaderTimeout != defaultTimeout {