agents(analyze-ci-failure): enhance network reset and macOS sandbox flake detection (#4037)

CI jobs on Buildkite runners occasionally encounter transient curl
network resets during runner bootstrapping and transient darwin-sandbox
disk I/O errors (errno 5) on macOS workers. Previously, the automated
CI failure analysis script did not match these error signatures,
leading to unclassified failures or treating them as codebase defects.

Add pattern matching and flake classification heuristics for curl
connection resets and runner I/O errors so they are recognized as
transient infrastructure flakes.

.agents/skills/analyze-ci-failure/scripts/analyze_ci_failure.py

diff --git a/.agents/skills/analyze-ci-failure/scripts/analyze_ci_failure.py b/.agents/skills/analyze-ci-failure/scripts/analyze_ci_failure.py
index 5708f6d..31d8dd7 100755
--- a/.agents/skills/analyze-ci-failure/scripts/analyze_ci_failure.py
+++ b/.agents/skills/analyze-ci-failure/scripts/analyze_ci_failure.py
@@ -114,6 +114,10 @@
                 "error waiting for container",
                 "error during connect:",
                 "user command error:",
+                "curl: (",
+                "connection reset by peer",
+                "input/output error",
+                "errno 5",
             ]
         ):
             if clean_line:
@@ -151,6 +155,19 @@
     ):
         is_flake = True
         flake_reason = "Buildkite agent / Docker runner infrastructure failure (dockerd disconnection / grpc context canceled / exit status 125). This is an infrastructure flake, not a codebase bug."
+    elif any(
+        "curl:" in e.lower()
+        or "recv failure: connection reset" in e.lower()
+        or "connection reset by peer" in e.lower()
+        for e in errors
+    ):
+        is_flake = True
+        flake_reason = "Network connection reset during runner bootstrap or artifact download (curl recv failure / connection reset). This is an infrastructure network flake."
+    elif any(
+        "input/output error" in e.lower() or "errno 5" in e.lower() for e in errors
+    ):
+        is_flake = True
+        flake_reason = "Transient runner host disk / Darwin sandbox I/O error (OSError: [Errno 5] Input/output error). This is an infrastructure flake, not a codebase defect."
 
     classification = (
         "⚡ **Classification**: **Infrastructure / Flake Issue** (Not a codebase bug)"