test: tolerate OpenClaw JSON envelope changes

test/e2e/lib/openclaw-json.sh

@@ -0,0 +1,90 @@
+#!/bin/bash
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+# Extract human-readable assistant text from `openclaw agent --json` output.
+# OpenClaw's JSON envelope has moved between result.payloads[] and top-level
+# payloads[]; keep E2E assertions focused on visible reply text instead of one
+# exact envelope shape. This also tolerates wrapper output before the JSON blob
+# but intentionally ignores metadata fields so IDs, durations, session names,
+# and model/provider details cannot satisfy reply assertions.
+parse_openclaw_agent_text() {
+  python3 -c '
+import json
+import sys
+
+raw = sys.stdin.read()
+if not raw.strip():
+    sys.exit(0)
+
+parts = []
+visited = set()
+
+TEXT_KEYS = {"text", "content", "reasoning_content"}
+CONTAINER_KEYS = {
+    "result", "payloads", "payload", "messages", "choices", "response",
+    "data", "output", "outputs", "items", "segments", "delta",
+}
+
+
+def add(value):
+    if isinstance(value, str) and value.strip():
+        parts.append(value.strip())
+
+
+def collect(value):
+    value_id = id(value)
+    if value_id in visited:
+        return
+    visited.add(value_id)
+
+    if isinstance(value, str):
+        add(value)
+        return
+    if isinstance(value, list):
+        for item in value:
+            collect(item)
+        return
+    if not isinstance(value, dict):
+        return
+
+    for key in TEXT_KEYS:
+        add(value.get(key))
+
+    # OpenAI-style choices can nest assistant text under message/delta objects.
+    for choice in value.get("choices") or []:
+        if isinstance(choice, dict):
+            collect(choice.get("message"))
+            collect(choice.get("delta"))
+            add(choice.get("text"))
+
+    for key in CONTAINER_KEYS:
+        if key in value:
+            collect(value[key])
+
+
+def collect_from_doc(doc):
+    if isinstance(doc, dict) and isinstance(doc.get("result"), dict):
+        collect(doc["result"])
+    else:
+        collect(doc)
+
+try:
+    collect_from_doc(json.loads(raw))
+except Exception:
+    decoder = json.JSONDecoder()
+    for idx, char in enumerate(raw):
+        if char != "{":
+            continue
+        try:
+            doc, _end = decoder.raw_decode(raw[idx:])
+        except Exception:
+            continue
+        before = len(parts)
+        collect_from_doc(doc)
+        if len(parts) > before:
+            break
+
+print("\n".join(parts))
+'
+}

test/e2e/test-bedrock-runtime-compatible-anthropic.sh

@@ -25,6 +25,8 @@ export NEMOCLAW_E2E_DEFAULT_TIMEOUT=2700
 SCRIPT_DIR_TIMEOUT="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
 # shellcheck source=test/e2e/e2e-timeout.sh
 . "${SCRIPT_DIR_TIMEOUT}/e2e-timeout.sh"
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+. "${SCRIPT_DIR_TIMEOUT}/lib/openclaw-json.sh"
 
 PASS=0
 FAIL=0
@@ -698,23 +700,7 @@ check_openclaw_agent_turn() {
     return
   fi
 
-  reply=$(printf '%s' "$raw" | python3 -c '
-import json
-import sys
-
-text = sys.stdin.read()
-for idx, char in enumerate(text):
-    if char != "{":
-        continue
-    try:
-        doc = json.loads(text[idx:])
-    except Exception:
-        continue
-    payloads = ((doc.get("result") or {}).get("payloads") or [])
-    parts = [p.get("text") for p in payloads if isinstance(p, dict) and isinstance(p.get("text"), str)]
-    print("\n".join(parts))
-    break
-' 2>/dev/null) || true
+  reply=$(printf '%s' "$raw" | parse_openclaw_agent_text 2>/dev/null) || true
 
   if [ "$rc" -eq 0 ] && grep -qi "PONG" <<<"$reply"; then
     pass "B8: OpenClaw agent completed a Bedrock-backed turn through inference.local"

test/e2e/test-brave-search-e2e.sh

@@ -33,6 +33,8 @@ export NEMOCLAW_E2E_DEFAULT_TIMEOUT=1800
 SCRIPT_DIR_TIMEOUT="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
 # shellcheck source=test/e2e/e2e-timeout.sh
 . "${SCRIPT_DIR_TIMEOUT}/e2e-timeout.sh"
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+. "${SCRIPT_DIR_TIMEOUT}/lib/openclaw-json.sh"
 
 PASS=0
 FAIL=0
@@ -306,19 +308,7 @@ check_real_brave_search_via_agent() {
     return
   fi
 
-  reply=$(printf '%s' "$raw" | python3 -c "
-import json, sys
-try:
-    doc = json.load(sys.stdin)
-except Exception:
-    sys.exit(0)
-result = doc.get('result') or {}
-parts = []
-for p in result.get('payloads') or []:
-    if isinstance(p, dict) and isinstance(p.get('text'), str):
-        parts.append(p['text'])
-print('\n'.join(parts))
-" 2>/dev/null) || true
+  reply=$(printf '%s' "$raw" | parse_openclaw_agent_text 2>/dev/null) || true
 
   # NVIDIA-related phrasing (nvidia, gpu, cuda, geforce) is overwhelmingly
   # likely in any legitimate top-1 web result for the query "NVIDIA".

test/e2e/test-full-e2e.sh

@@ -69,43 +69,8 @@ except Exception as e:
 "
 }
 
-parse_openclaw_agent_text() {
-  python3 -c '
-import json
-import sys
-
-try:
-    doc = json.load(sys.stdin)
-except Exception:
-    sys.exit(0)
-
-parts = []
-
-def collect(value):
-    if isinstance(value, str):
-        if value.strip():
-            parts.append(value)
-        return
-    if isinstance(value, list):
-        for item in value:
-            collect(item)
-        return
-    if not isinstance(value, dict):
-        return
-
-    for key in ("text", "content", "reasoning_content", "message"):
-        found = value.get(key)
-        if isinstance(found, str) and found.strip():
-            parts.append(found)
-
-    for key in ("payloads", "payload", "messages", "choices", "result", "response", "data", "output"):
-        if key in value:
-            collect(value[key])
-
-collect(doc.get("result", doc))
-print("\n".join(parts))
-'
-}
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/lib/openclaw-json.sh"
 
 # Determine repo root
 if [ -d /workspace ] && [ -f /workspace/install.sh ]; then
@@ -409,8 +374,8 @@ fi
 #     routeLogsToStderr() (openclaw/src/commands/agent-via-gateway.ts:57),
 #     so stdout is a clean JSON envelope; prompt-echo on stderr cannot
 #     pollute the assertion.
-#   * Asserts on the model's reply text inside `result.payloads[].text`,
-#     not on the merged stdout/stderr.
+#   * Asserts on parsed model reply text from the JSON envelope, not on
+#     the merged stdout/stderr or a single brittle envelope shape.
 #   * The expected token (the integer 42) is not a literal substring of the
 #     prompt, so an error path that quoted the prompt back cannot satisfy
 #     the grep.

test/e2e/test-launchable-smoke.sh

@@ -52,6 +52,8 @@ export NEMOCLAW_E2E_DEFAULT_TIMEOUT=1800
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
 # shellcheck source=test/e2e/e2e-timeout.sh
 source "${SCRIPT_DIR}/e2e-timeout.sh"
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+source "${SCRIPT_DIR}/lib/openclaw-json.sh"
 
 PASS=0
 FAIL=0
@@ -522,19 +524,7 @@ if openshell sandbox ssh-config "$SANDBOX_NAME" >"$ssh_config" 2>/dev/null; then
 fi
 rm -f "$ssh_config"
 
-agent_reply=$(echo "$agent_response" | python3 -c "
-import json, sys
-try:
-    doc = json.load(sys.stdin)
-except Exception:
-    sys.exit(0)
-result = doc.get('result') or {}
-parts = []
-for p in result.get('payloads') or []:
-    if isinstance(p, dict) and isinstance(p.get('text'), str):
-        parts.append(p['text'])
-print('\n'.join(parts))
-" 2>/dev/null) || true
+agent_reply=$(printf '%s' "$agent_response" | parse_openclaw_agent_text 2>/dev/null) || true
 
 if grep -qE "(^|[^0-9])42([^0-9]|$)" <<<"$agent_reply"; then
   pass "[LIVE] openclaw agent: model answered 6×7=42 through openclaw → inference.local"

test/e2e/test-messaging-compatible-endpoint.sh

@@ -37,6 +37,8 @@ export NEMOCLAW_E2E_DEFAULT_TIMEOUT=1800
 SCRIPT_DIR_TIMEOUT="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
 # shellcheck source=test/e2e/e2e-timeout.sh
 . "${SCRIPT_DIR_TIMEOUT}/e2e-timeout.sh"
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+. "${SCRIPT_DIR_TIMEOUT}/lib/openclaw-json.sh"
 
 PASS=0
 FAIL=0
@@ -525,19 +527,7 @@ check_openclaw_agent_turn() {
     return
   fi
 
-  reply=$(printf '%s' "$raw" | python3 -c "
-import json, sys
-try:
-    doc = json.load(sys.stdin)
-except Exception:
-    sys.exit(0)
-result = doc.get('result') or {}
-parts = []
-for p in result.get('payloads') or []:
-    if isinstance(p, dict) and isinstance(p.get('text'), str):
-        parts.append(p['text'])
-print('\n'.join(parts))
-" 2>/dev/null) || true
+  reply=$(printf '%s' "$raw" | parse_openclaw_agent_text 2>/dev/null) || true
 
   if [ "$rc" -eq 0 ] && printf '%s' "$reply" | grep -qi "PONG"; then
     pass "C8: openclaw agent completed turn via compatible endpoint (http-proxy-fix.js FORWARD-mode path exercised)"

test/e2e/test-openclaw-inference-switch.sh

@@ -274,20 +274,7 @@ check_openclaw_agent_turn() {
     2>/dev/null) || rc=$?
   rm -f "$ssh_config"
 
-  reply=$(printf '%s' "$raw" | python3 -c '
-import json
-import sys
-try:
-    doc = json.load(sys.stdin)
-except Exception:
-    sys.exit(0)
-result = doc.get("result") or {}
-parts = []
-for payload in result.get("payloads") or []:
-    if isinstance(payload, dict) and isinstance(payload.get("text"), str):
-        parts.append(payload["text"])
-print("\n".join(parts))
-' 2>/dev/null) || true
+  reply=$(printf '%s' "$raw" | parse_openclaw_agent_text 2>/dev/null) || true
 
   if [ "$rc" -eq 0 ] && grep -qE '(^|[^0-9])42([^0-9]|$)' <<<"$reply"; then
     pass "OpenClaw agent answered through the switched inference route"
@@ -306,6 +293,8 @@ else
 fi
 
 E2E_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+. "${E2E_DIR}/lib/openclaw-json.sh"
 SANDBOX_NAME="${NEMOCLAW_SANDBOX_NAME:-e2e-openclaw-inference-switch}"
 SWITCH_PROVIDER="${NEMOCLAW_SWITCH_PROVIDER:-nvidia-prod}"
 SWITCH_MODEL="${NEMOCLAW_SWITCH_MODEL:-z-ai/glm-5.1}"

test/e2e/test-sandbox-operations.sh

@@ -24,6 +24,8 @@ export NEMOCLAW_E2E_DEFAULT_TIMEOUT=1800
 SCRIPT_DIR_TIMEOUT="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
 # shellcheck source=test/e2e/e2e-timeout.sh
 source "${SCRIPT_DIR_TIMEOUT}/e2e-timeout.sh"
+# shellcheck source=test/e2e/lib/openclaw-json.sh
+source "${SCRIPT_DIR_TIMEOUT}/lib/openclaw-json.sh"
 
 # ── Config ───────────────────────────────────────────────────────────────────
 SANDBOX_A="test-sbx-a"
@@ -353,8 +355,8 @@ test_sbx_01_list_sandboxes() {
 #      the prompt, so an error path that quoted the prompt back cannot
 #      false-positive the grep — which is what masked the openclaw 4.9
 #      SSRF regression from the prior `Say exactly: HELLO_E2E` assertion.
-#   3. Asserts on `result.payloads[].text` from the JSON envelope, not on
-#      merged stdout/stderr.
+#   3. Asserts on parsed model reply text from the JSON envelope, not on
+#      merged stdout/stderr or a single brittle envelope shape.
 #   4. Pins `--thinking off` so the first-turn smoke contract is not delayed
 #      by model-catalog inferred reasoning defaults.
 test_sbx_02_connect_chat() {
@@ -384,19 +386,7 @@ test_sbx_02_connect_chat() {
   rm -f "$ssh_cfg"
 
   local reply
-  reply=$(echo "$raw" | python3 -c "
-import json, sys
-try:
-    doc = json.load(sys.stdin)
-except Exception:
-    sys.exit(0)
-result = doc.get('result') or {}
-parts = []
-for p in result.get('payloads') or []:
-    if isinstance(p, dict) and isinstance(p.get('text'), str):
-        parts.append(p['text'])
-print('\n'.join(parts))
-" 2>/dev/null) || true
+  reply=$(printf '%s' "$raw" | parse_openclaw_agent_text 2>/dev/null) || true
 
   if [[ -n "$reply" ]] && echo "$reply" | grep -qE "(^|[^0-9])42([^0-9]|$)"; then
     pass "TC-SBX-02: Agent computed 6×7=42 through openclaw → inference.local"