diff --git a/docs/hermes-openclaw-a2a-current-state.json b/docs/hermes-openclaw-a2a-current-state.json
new file mode 100644
index 0000000000000..70093a6d36fc6
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-current-state.json
@@ -0,0 +1,276 @@
+{
+  "repo_a2a_items": [
+    {
+      "path": "/.hermes/hermes-agent/docs/hermes-openclaw-a2a-task-plan-v1.0.0.md",
+      "type": "file",
+      "size": 8876
+    },
+    {
+      "path": "/.hermes/hermes-agent/docs/hermes-openclaw-a2a-task-plan-v1.1.0.md",
+      "type": "file",
+      "size": 13412
+    },
+    {
+      "path": "/.hermes/hermes-agent/docs/hermes-openclaw-a2a-v240-recurring-cron-template.md",
+      "type": "file",
+      "size": 3623
+    },
+    {
+      "path": "/.hermes/hermes-agent/docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md",
+      "type": "file",
+      "size": 3677
+    },
+    {
+      "path": "/.hermes/hermes-agent/docs/hermes-openclaw-a2a-worklog-and-architecture.md",
+      "type": "file",
+      "size": 9349
+    }
+  ],
+  "home_a2a_items": [
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-adaptive-dispatch-authorization-acceptance.md",
+      "type": "file",
+      "size": 3169
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-role-bound-dispatch-plan.md",
+      "type": "file",
+      "size": 1976
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v067.md",
+      "type": "file",
+      "size": 4105
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v070-callback-loop.md",
+      "type": "file",
+      "size": 3143
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v071-guarded-callback-send.md",
+      "type": "file",
+      "size": 3508
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v072-one-shot-runner.md",
+      "type": "file",
+      "size": 3331
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v100-bidirectional-minimal-loop.md",
+      "type": "file",
+      "size": 5245
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v110-failure-paths.md",
+      "type": "file",
+      "size": 3679
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v120-explicit-queue-runner.md",
+      "type": "file",
+      "size": 2235
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v120-local-queue-runner.md",
+      "type": "file",
+      "size": 4802
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v120-v130-queue-runners.md",
+      "type": "file",
+      "size": 4812
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v120-v200-queue-cli-cron.md",
+      "type": "file",
+      "size": 6738
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v130-mixed-queue-isolation.md",
+      "type": "file",
+      "size": 2615
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v140-queue-cli-schema.md",
+      "type": "file",
+      "size": 2774
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v210-dispatch-compact-callback.md",
+      "type": "file",
+      "size": 2744
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v220-compact-cron-callback.md",
+      "type": "file",
+      "size": 2302
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v230-compact-cronjob-boundary.md",
+      "type": "file",
+      "size": 2345
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v231-cron-scheduler-run-verification.md",
+      "type": "file",
+      "size": 1980
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v232-cron-runtime-monitor.md",
+      "type": "file",
+      "size": 2069
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v240-manual-cron-monitor-race.md",
+      "type": "file",
+      "size": 2670
+    },
+    {
+      "path": "/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/hermes-openclaw-a2a-v260-controlled-manual-cron-run.md",
+      "type": "file",
+      "size": 2679
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-platform-delivery-and-channel-routing/references/hermes-openclaw-a2a-receipt-callback.md",
+      "type": "file",
+      "size": 5256
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-a2a-gateway-live-install.md",
+      "type": "file",
+      "size": 5559
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-a2a-webhooks-bridge.md",
+      "type": "file",
+      "size": 4397
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-a2a-webhooks-planning.md",
+      "type": "file",
+      "size": 3278
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-node22-direct-upgrade-first.md",
+      "type": "file",
+      "size": 2351
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-v062-readiness-assessment.md",
+      "type": "file",
+      "size": 3229
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-v066-authenticated-a2a-ping.md",
+      "type": "file",
+      "size": 3203
+    },
+    {
+      "path": "/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/hermes-openclaw-vaultwarden-ssh-node22-readiness.md",
+      "type": "file",
+      "size": 4140
+    }
+  ],
+  "scripts": [
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_v120_queue_runner.py",
+      "size": 10682
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_v100_orchestrator.py",
+      "size": 13900
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_queue_cron_v220.sh",
+      "size": 500
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_queue_summary.py",
+      "size": 2931
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_cron_monitor.py",
+      "size": 8590
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_v130_mixed_queue_runner.py",
+      "size": 15179
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_queue_cron_v200.sh",
+      "size": 266
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_v110_failure_tests.py",
+      "size": 9957
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/hermes_openclaw_queue.py",
+      "size": 20988
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/__pycache__/hermes_openclaw_v110_failure_tests.cpython-311.pyc",
+      "size": 14784
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/__pycache__/hermes_openclaw_cron_monitor.cpython-311.pyc",
+      "size": 14537
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/__pycache__/hermes_openclaw_v130_mixed_queue_runner.cpython-311.pyc",
+      "size": 24773
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/__pycache__/hermes_openclaw_v100_orchestrator.cpython-311.pyc",
+      "size": 28450
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/__pycache__/hermes_openclaw_v120_queue_runner.cpython-311.pyc",
+      "size": 17464
+    },
+    {
+      "path": "/.hermes/hermes-agent/scripts/__pycache__/hermes_openclaw_queue.cpython-311.pyc",
+      "size": 46266
+    },
+    {
+      "path": "/root/.hermes/scripts/openclaw_queue_cron_v200.sh",
+      "size": 266
+    },
+    {
+      "path": "/root/.hermes/scripts/openclaw_queue_cron_v220.sh",
+      "size": 500
+    }
+  ],
+  "evidence_dirs_sample": [
+    {
+      "path": "/.hermes/hermes-agent/optional-skills/migration/openclaw-migration",
+      "mtime": 1779165541.222174
+    },
+    {
+      "path": "/.hermes/hermes-agent/optional-skills/migration/openclaw-migration/scripts",
+      "mtime": 1779165541.222174
+    }
+  ],
+  "git_status": {
+    "code": 0,
+    "out": "## main...origin/main [ahead 2, behind 973]\n M agent/conversation_loop.py\n M gateway/platforms/base.py\n M gateway/platforms/feishu.py\n M gateway/run.py\n M gateway/stream_consumer.py\n M hermes_cli/kanban_db.py\n M tools/approval.py\n?? curation-consolidation-report.md\n?? docs/dispatch-strategy-v2.1.0.md\n?? docs/execution-v1.0.0.md\n?? docs/execution-v1.1.0.md\n?? docs/execution-v1.2.0.md\n?? docs/execution-v1.3.0.md\n?? docs/execution-v1.4.0.md\n?? docs/execution-v1.6.0.md\n?? docs/execution-v1.7.0.md\n?? docs/execution-v1.9.0.md\n?? docs/execution-v2.0.0.md\n?? docs/execution-v2.2.0.md\n?? docs/execution-v2.3.0.md\n?? docs/execution-v2.3.1.md\n?? docs/execution-v2.3.2.md\n?? docs/hermes-openclaw-a2a-task-plan-v1.0.0.md\n?? docs/hermes-openclaw-a2a-task-plan-v1.1.0.md\n?? docs/hermes-openclaw-a2a-v240-recurring-cron-template.md\n?? docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md\n?? docs/hermes-openclaw-a2a-worklog-and-architecture.md\n?? docs/validation-v1.0.0.md\n?? docs/validation-v1.1.0.md\n?? docs/validation-v1.2.0.md\n?? docs/validation-v1.3.0.md\n?? docs/validation-v1.4.0.md\n?? docs/validation-v1.6.0.md\n?? docs/validation-v1.7.0.md\n?? docs/validation-v1.9.0.md\n?? docs/validation-v2.0.0.md\n?? docs/validation-v2.2.0.md\n?? docs/validation-v2.3.0.md\n?? docs/validation-v2.3.1.md\n?? docs/validation-v2.3.2.md\n?? gateway/platforms/feishu.py.bak\n?? gateway/platforms/feishu.py.bak.20260522_171805\n?? gateway/platforms/feishu.py.bak.20260522_171812\n?? gateway/platforms/feishu.py.bak2\n?? gateway/stream_consumer.py.bak\n?? scripts/hermes_openclaw_cron_monitor.py\n?? scripts/hermes_openclaw_queue.py\n?? scripts/hermes_openclaw_queue_cron_v200.sh\n?? scripts/hermes_openclaw_queue_cron_v220.sh\n?? scripts/hermes_openclaw_queue_summary.py\n?? scripts/hermes_openclaw_v100_orchestrator.py\n?? scripts/hermes_openclaw_v110_failure_tests.py\n?? scripts/hermes_openclaw_v120_queue_runner.py\n?? scripts/hermes_openclaw_v130_mixed_queue_runner.py\n?? skill-clusters-analysis.json\n"
+  },
+  "cron_list": {
+    "code": 0,
+    "out": "No scheduled jobs.\nCreate one with 'hermes cron create ...' or the /cron command in chat.\n"
+  },
+  "gateway_status": {
+    "code": 0,
+    "out": "me__, str(e))\\\\nPY2'\\\\''\\n}\\nfor k,cmd in cmds.items():\\n    try:\\n        r=subprocess.run(cmd,shell=True,stdout=subprocess.PIPE,stderr=subprocess.STDOUT,text=True,timeout=30)\\n        report[k]={'\\\\''code'\\\\'':r.returncode,'\\\\''out'\\\\'':r.stdout[-4000:]}\\n    except Exception as e:\\n        report[k]={'\\\\''error'\\\\'':repr(e)}\\nout=Path('\\\\''/.hermes/hermes-agent/docs/hermes-openclaw-a2a-current-state.json'\\\\'')\\nout.write_text(json.dumps(report,ensure_ascii=False,indent=2),encoding='\\\\''utf-8'\\\\'')\\nprint(out)\\nprint(out.stat().st_size)\\nprint('\\\\''docs'\\\\'', len(report.get('\\\\''repo_a2a_items'\\\\'',[])), '\\\\''home_items'\\\\'', len(report.get('\\\\''home_a2a_items'\\\\'',[])), '\\\\''scripts'\\\\'', len(scripts), '\\\\''evidence'\\\\'', len(report['\\\\''evidence_dirs_sample'\\\\'']))\\nprint('\\\\''port'\\\\'', report['\\\\''port_18789'\\\\''])\\nprint('\\\\''agent_card_first'\\\\'', report['\\\\''agent_card'\\\\'']['\\\\''out'\\\\''][:500] if '\\\\''out'\\\\'' in report['\\\\''agent_card'\\\\''] else report['\\\\''agent_card'\\\\''])\\nPY'\\n__hermes_ec=\\$?\\nexport -p > /tmp/hermes-snap-5a3c29d16380.sh 2>/dev/null || true\\npwd -P > /tmp/hermes-cwd-5a3c29d16380.txt 2>/dev/null || true\\nprintf '\\\\n__HERMES_CWD_5a3c29d16380__%s__HERMES_CWD_5a3c29d16380__\\\\n' \\\"\\$(pwd -P)\\\"\\nexit \\$__hermes_ec\"\n             ├─10427 python3 -\n             ├─10440 /bin/sh -c \"hermes gateway status\"\n             ├─10441 /.hermes/hermes-agent/venv/bin/python3 /root/.local/bin/hermes gateway status\n             └─10453 systemctl status hermes-gateway --no-pager\n\nMay 29 21:35:10 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool terminal returned error (0.28s): {\"output\": \"/usr/bin/bash: line 24: examples/v2.2.0/manual-v220-stdout.md: No such file or directory\", \"exit_code\": 1, \"error\": null}\nMay 29 21:37:18 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool skill_manage returned error (0.00s): {\"success\": false, \"error\": \"File must be under one of: assets, references, scripts, templates. Got: 'SKILL.md'\"}\nMay 29 21:37:48 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool cronjob returned error (0.00s): {\"error\": \"Invalid schedule 'once at 2099-01-01 00:00'. Use:\\n  - Duration: '30m', '2h', '1d' (one-shot)\\n  - Interval: 'every 30m', 'every 2h' (recurring)\\n  - Cron: '0 9 * * *' (cron expression)\\n\nMay 29 21:38:45 Hermes-248 python[1014]: WARNING gateway.platforms.qqbot.adapter: [QQBot:1903759289] WebSocket closed: code=4009 reason=Session timed out\nMay 29 21:40:25 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool skill_manage returned error (0.00s): {\"success\": false, \"error\": \"File must be under one of: assets, references, scripts, templates. Got: 'SKILL.md'\"}\nMay 29 21:48:24 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool skill_manage returned error (0.00s): {\"success\": false, \"error\": \"File must be under one of: assets, references, scripts, templates. Got: 'SKILL.md'\"}\nMay 29 22:05:03 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool terminal returned error (10.68s): {\"output\": \"\", \"exit_code\": 2, \"error\": null}\nMay 29 22:05:53 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool terminal returned error (20.68s): {\"output\": \"monitor_exists True 2069\\nok False\\ndiagnosis process still running\\nexecution_observed False\\nnew_evidence ['cron-run-20260529T140519Z']\\nprocess_running_at_end True\\ninitial_count 3\\nfin\nMay 29 22:08:47 Hermes-248 python[1014]: WARNING gateway.platforms.qqbot.adapter: [QQBot:1903759289] WebSocket closed: code=4009 reason=Session timed out\nMay 29 22:16:08 Hermes-248 python[1014]: WARNING agent.tool_executor: Tool skill_manage returned error (0.00s): {\"success\": false, \"error\": \"File must be under one of: assets, references, scripts, templates. Got: 'SKILL.md'\"}\n⚠ Installing gateway service to run as root.\n    This is fine for LXC/container environments but not recommended on bare-metal hosts.\n✓ System gateway service is running\nConfigured to run as: root\n✓ System service starts at boot without requiring systemd linger\n"
+  },
+  "port_18789": {
+    "code": 0,
+    "out": "open\n"
+  },
+  "agent_card": {
+    "code": 0,
+    "out": "HTTPError HTTP Error 404: Not Found\n"
+  }
+}
\ No newline at end of file
diff --git a/docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md b/docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md
new file mode 100644
index 0000000000000..ade5a95a2e7dc
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md
@@ -0,0 +1,61 @@
+# Hermes ↔ OpenClaw A2A v2.6.x File Inventory
+
+## 生成时间
+本文件由 v2.6.4 runbook closure 阶段生成，用于记录当前工作区内 v2.6.x 相关新增/修改文件。
+
+## Git 工作区清单
+```text
+M docs/hermes-openclaw-a2a-worklog-and-architecture.md
+?? docs/hermes-openclaw-a2a-live-validation-v2.6.0.md
+?? docs/hermes-openclaw-a2a-phase6-review-v2.6.0.md
+?? docs/hermes-openclaw-a2a-review-v2.6.0.md
+?? docs/hermes-openclaw-a2a-review-v2.6.1.md
+?? docs/hermes-openclaw-a2a-review-v2.6.2.md
+?? docs/hermes-openclaw-a2a-review-v2.6.3.md
+?? docs/hermes-openclaw-a2a-task-plan-v2.6.0.md
+?? docs/hermes-openclaw-a2a-task-plan-v2.6.1.md
+?? docs/hermes-openclaw-a2a-task-plan-v2.6.2.md
+?? docs/hermes-openclaw-a2a-task-plan-v2.6.3.md
+?? docs/hermes-openclaw-a2a-validation-v2.6.0.md
+?? docs/hermes-openclaw-a2a-validation-v2.6.1.md
+?? docs/hermes-openclaw-a2a-validation-v2.6.2.md
+?? docs/hermes-openclaw-a2a-validation-v2.6.3.md
+?? scripts/hermes_openclaw_v260_two_worker.py
+?? scripts/validate_a2a_v260_evidence.py
+?? scripts/validate_a2a_v260_mock.py
+?? scripts/validate_a2a_v260_negative.py
+?? scripts/verify_a2a_v263_chain.py
+```
+
+## v2.6.x 核心脚本
+- `scripts/hermes_openclaw_v260_two_worker.py`：v2.6.0 two-worker runner，支持 dry-run 与受控 live 模式。
+- `scripts/validate_a2a_v260_mock.py`：v2.6.0 mock fixture validator。
+- `scripts/validate_a2a_v260_evidence.py`：v2.6.1 positive evidence validator。
+- `scripts/validate_a2a_v260_negative.py`：v2.6.2 negative failure-path validator。
+- `scripts/verify_a2a_v263_chain.py`：v2.6.3 一键 verify chain。
+
+## v2.6.x 核心文档
+- `docs/hermes-openclaw-a2a-task-plan-v2.6.0.md`
+- `docs/hermes-openclaw-a2a-validation-v2.6.0.md`
+- `docs/hermes-openclaw-a2a-review-v2.6.0.md`
+- `docs/hermes-openclaw-a2a-live-validation-v2.6.0.md`
+- `docs/hermes-openclaw-a2a-phase6-review-v2.6.0.md`
+- `docs/hermes-openclaw-a2a-task-plan-v2.6.1.md`
+- `docs/hermes-openclaw-a2a-validation-v2.6.1.md`
+- `docs/hermes-openclaw-a2a-review-v2.6.1.md`
+- `docs/hermes-openclaw-a2a-task-plan-v2.6.2.md`
+- `docs/hermes-openclaw-a2a-validation-v2.6.2.md`
+- `docs/hermes-openclaw-a2a-review-v2.6.2.md`
+- `docs/hermes-openclaw-a2a-task-plan-v2.6.3.md`
+- `docs/hermes-openclaw-a2a-validation-v2.6.3.md`
+- `docs/hermes-openclaw-a2a-review-v2.6.3.md`
+- `docs/hermes-openclaw-a2a-runbook-v2.6.4.md`
+- `docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md`
+
+- `docs/hermes-openclaw-a2a-precommit-audit-v2.6.x.md`
+
+## 当前边界
+- v2.6.0 包含一次历史受控 live two-worker evidence。
+- v2.6.1/v2.6.2/v2.6.3/v2.6.4 不新增 live call。
+- 当前未启用 cron / daemon / webhook / reverse loop。
+- 当前没有提交 git commit。
diff --git a/docs/hermes-openclaw-a2a-live-validation-v2.6.0.md b/docs/hermes-openclaw-a2a-live-validation-v2.6.0.md
new file mode 100644
index 0000000000000..c034b36a70bef
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-live-validation-v2.6.0.md
@@ -0,0 +1,128 @@
+# Hermes ↔ OpenClaw A2A v2.6.0 Live Validation
+
+## 结论
+
+PASS with boundary：v2.6.0 已完成从本地 mock 到 live two-worker 的受控样例验证。
+
+Validation marker：`A2A_V260_LIVE_TWO_WORKER_OK`
+
+## 执行范围
+
+本次执行严格对应 v2.6.0 实施顺序：
+
+1. 计划落盘并回读验证：已完成。
+2. 总入口文档指向 v2.6.0：已完成。
+3. dispatch envelope / receipt / acceptance fixture：已完成。
+4. 本地 mock schema validation：已完成。
+5. live two-worker A2A 样例：已完成。
+6. Hermes 回收 evidence 并生成 acceptance report：已完成。
+
+未进入第 7 步：未固化为 queue CLI 或 Kanban/Swarm 模板。
+
+## 实际执行命令
+
+```bash
+cd /.hermes/hermes-agent
+python3 -m py_compile scripts/hermes_openclaw_v260_two_worker.py
+python3 scripts/hermes_openclaw_v260_two_worker.py \
+  --out-dir examples/v2.6.0/live-two-worker
+```
+
+## 真实执行结果
+
+```json
+{
+  "ok": true,
+  "run_id": "a2a-v260-two-worker-20260530T091052Z",
+  "dry_run": false,
+  "receipt_count": 2,
+  "accepted_count": 2,
+  "overall": "accepted_with_boundary",
+  "secret_scan_ok": true,
+  "out_dir": "examples/v2.6.0/live-two-worker",
+  "compact_summary": "结论：A2A v2.6.0 two-worker 样例 accepted_with_boundary\nrun_id: a2a-v260-two-worker-20260530T091052Z\nOpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0\n证据：examples/v2.6.0/live-two-worker\n边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息\n下一步：只在明确要求后，才固化为 queue CLI 或 Kanban/Swarm 模板\n"
+}
+```
+
+## 关键证据
+
+- live evidence dir：`examples/v2.6.0/live-two-worker`
+- dry-run evidence dir：`examples/v2.6.0/dry-run-two-worker`
+- live readiness：`examples/v2.6.0/live-two-worker/readiness.json`
+- Worker A receipt：`examples/v2.6.0/live-two-worker/a2a-v260-worker-readiness-receipt.json`
+- Worker B receipt：`examples/v2.6.0/live-two-worker/a2a-v260-worker-review-receipt.json`
+- Acceptance report：`examples/v2.6.0/live-two-worker/acceptance-report.json`
+- Execution summary：`examples/v2.6.0/live-two-worker/execution-summary.json`
+- Compact summary：`examples/v2.6.0/live-two-worker/compact-summary.md`
+
+## Live readiness
+
+`readiness.json` 记录：
+
+- `agent_card_http_status=200`
+- `agent_card_live=true`
+- `credential_loaded=true`
+- `credential_recorded=false`
+
+这表示 live endpoint 可达，授权凭据仅用于本次请求，不写入证据文件。
+
+## Worker 验收
+
+Acceptance report 中两个子任务均为 accepted：
+
+- `a2a-v260-worker-readiness`：accepted
+- `a2a-v260-worker-review`：accepted
+
+两个 receipt 均满足：
+
+- `schema_version=a2a-worker-receipt-v1`
+- `ok=true`
+- `http_status=200`
+- `state=completed`
+- `auth.token_recorded=false`
+- marker 在对应 evidence 文件中存在
+
+## Side effects
+
+Acceptance report 记录：
+
+```json
+{
+  "gateway_restart": false,
+  "openclaw_restart": false,
+  "cron_enabled": false,
+  "platform_send": false,
+  "live_a2a_call": true
+}
+```
+
+解释：
+
+- `live_a2a_call=true`：本轮确实执行了 live Hermes → OpenClaw A2A JSON-RPC `message/send`。
+- `gateway_restart=false`：未重启 Hermes gateway。
+- `openclaw_restart=false`：未重启 OpenClaw。
+- `cron_enabled=false`：未启用 cron。
+- `platform_send=false`：未做额外平台外发。
+
+## Secret scan
+
+Acceptance report 记录：
+
+```json
+{
+  "ok": true,
+  "token_recorded": false,
+  "forbidden_literals_found": []
+}
+```
+
+本地回读扫描也显示：`violations=[]`。
+
+## 边界
+
+本轮通过的是受控 live two-worker 样例，不是开放式自治协作：
+
+- 未启用 OpenClaw → Hermes 反向调用。
+- 未启用 daemon / webhook / recurring automation。
+- 未将该样例固化为 queue CLI 或 Kanban/Swarm 模板。
+- Hermes 仍是 controller 和 final acceptor。
diff --git a/docs/hermes-openclaw-a2a-openclaw-gateway-blocking-response-fix.md b/docs/hermes-openclaw-a2a-openclaw-gateway-blocking-response-fix.md
new file mode 100644
index 0000000000000..0a2aa94ce921a
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-openclaw-gateway-blocking-response-fix.md
@@ -0,0 +1,135 @@
+# OpenClaw A2A gateway blocking response patch note
+
+## Context
+
+During Hermes ↔ OpenClaw A2A v2.6 live validation, Hermes could deliver tasks to OpenClaw and OpenClaw executed them, but the blocking `message/send` HTTP call timed out before receiving the final JSON-RPC result.
+
+Layered status before the fix:
+
+- Hermes → OpenClaw task delivery: passed
+- OpenClaw task execution: passed
+- OpenClaw log/audit marker: passed
+- OpenClaw → Hermes blocking JSON-RPC response: failed / timed out
+- two-worker live acceptance: blocked
+
+## Root cause
+
+OpenClaw gateway returned the initial `agent` RPC as `accepted`, while the final agent output arrived later as a WebSocket `event=agent` frame without the original request id.
+
+The existing `GatewayRpcConnection.handleMessage()` path ignored event frames after `connect.challenge`, so the pending `expectFinal` agent request was never resolved. Hermes therefore waited until the blocking HTTP client timed out.
+
+## Runtime fix applied on OpenClaw 247
+
+Runtime repo:
+
+```text
+/root/.openclaw/extensions/a2a-gateway
+```
+
+Changed file:
+
+```text
+src/executor.ts
+```
+
+Patch class:
+
+- preserve `connect.challenge` behavior;
+- for `event=agent`, inspect `payload.result.payloads`;
+- only resolve when the event contains real agent text/media payload content;
+- if exactly one pending `expectFinal` `agent` request exists, resolve it with the event payload;
+- ignore non-content agent events to avoid resolving too early.
+
+Current uncommitted runtime diff:
+
+```diff
+diff --git a/src/executor.ts b/src/executor.ts
+index d5ab86c..05334c4 100644
+--- a/src/executor.ts
++++ b/src/executor.ts
+@@ -921,6 +921,24 @@ export class GatewayRpcConnection {
+         if (nonce && this.connectChallengeResolver) {
+           this.connectChallengeResolver(nonce);
+         }
++        return;
++      }
++
++      if (frame.event === "agent") {
++        const eventPayload = asObject(frame.payload);
++        const eventResult = asObject(eventPayload?.result);
++        const eventPayloads = Array.isArray(eventResult?.payloads) ? eventResult.payloads : [];
++        const hasAgentContent = eventPayloads.some((entry) => Boolean(extractAgentPayloadText(entry)) || extractMediaUrlsFromPayload(entry).length > 0);
++        if (hasAgentContent) {
++          const pendingEntries = Array.from(this.pending.entries())
++            .filter(([, entry]) => entry.expectFinal && entry.method === "agent");
++          if (pendingEntries.length === 1) {
++            const [pendingId, pending] = pendingEntries[0];
++            this.pending.delete(pendingId);
++            clearTimeout(pending.timer);
++            pending.resolve(frame.payload);
++          }
++        }
+       }
+       return;
+     }
+```
+
+## Verification
+
+TypeScript check on OpenClaw 247:
+
+```text
+cd /root/.openclaw/extensions/a2a-gateway
+npx tsc --noEmit --pretty false
+# passed
+```
+
+Gateway service:
+
+```text
+systemctl is-active openclaw-gateway.service
+active
+```
+
+Latest live Hermes evidence:
+
+```text
+examples/v2.6.0/live-final-fix2-20260530T135309Z
+```
+
+Runner result:
+
+```json
+{
+  "ok": true,
+  "run_id": "a2a-v260-two-worker-20260530T135309Z",
+  "dry_run": false,
+  "receipt_count": 2,
+  "accepted_count": 2,
+  "overall": "accepted_with_boundary",
+  "secret_scan_ok": true
+}
+```
+
+Evidence validator:
+
+```text
+ok: true
+receipt_count: 2
+accepted_count: 2
+overall: accepted_with_boundary
+```
+
+Secret scan:
+
+```text
+FINAL_SECRET_HIT_COUNT 0
+FINAL_SECRET_HITS []
+```
+
+## Side effects
+
+- OpenClaw gateway was restarted during diagnosis and after the fix.
+- No Hermes config was changed.
+- No cron, daemon, webhook, reverse loop, or platform send was enabled.
+- The runtime OpenClaw patch is currently local to `192.168.31.247`; it still needs to be committed/upstreamed in the OpenClaw A2A gateway project if long-term persistence across plugin updates is required.
diff --git a/docs/hermes-openclaw-a2a-phase6-assessment.json b/docs/hermes-openclaw-a2a-phase6-assessment.json
new file mode 100644
index 0000000000000..eb09c81b77086
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-phase6-assessment.json
@@ -0,0 +1,32 @@
+{
+  "phase": "Phase 6",
+  "created_at": "2026-05-29T14:29:26.859791+00:00",
+  "assessment": "bounded two-turn Hermes-controlled A2A loop is already verified via Phase 1 live run; autonomous reverse routing is not enabled and should not be enabled by default.",
+  "decision": "accepted_with_boundary",
+  "enter_reverse_implementation": false,
+  "reasons": [
+    "Hermes remains controller and final acceptor.",
+    "Current live proof covers Hermes -> OpenClaw -> Hermes receipt/summary for two turns.",
+    "Failure paths and queue runner are verified.",
+    "Cron template exists and is paused.",
+    "OpenClaw-initiated reverse calls / autonomous loops still need explicit anti-loop, auth, target routing, and user approval before implementation."
+  ],
+  "required_before_reverse": [
+    "max_rounds and loop guard",
+    "dedupe ledger for reverse callbacks",
+    "strict allowlist of callable actions",
+    "origin delivery dry-run for reverse path",
+    "failure/unsafe classification for reverse requests"
+  ],
+  "evidence": {
+    "phase0_state_json": "docs/hermes-openclaw-a2a-current-state.json",
+    "phase1_evidence": "examples/live-phase1",
+    "phase3_evidence": "examples/v1.1.0",
+    "phase4_evidence": "examples/live-phase4-run",
+    "phase5_cron_job": "170596628792 paused",
+    "phase5_docs": [
+      "docs/hermes-openclaw-a2a-v240-recurring-cron-template.md",
+      "docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md"
+    ]
+  }
+}
\ No newline at end of file
diff --git a/docs/hermes-openclaw-a2a-phase6-assessment.md b/docs/hermes-openclaw-a2a-phase6-assessment.md
new file mode 100644
index 0000000000000..b0720bf0c821d
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-phase6-assessment.md
@@ -0,0 +1,60 @@
+# Hermes ↔ OpenClaw A2A Phase 6 Assessment
+
+## 结论
+
+验收分类：`accepted_with_boundary`
+
+当前已经验证的是 **Hermes 控制的 bounded two-turn A2A 对话**：Hermes 派发第 1 轮，验收 receipt；再基于第 1 轮摘要派发第 2 轮，继续验收 receipt，并由 Hermes 承载当前通道回流。
+
+当前**不默认进入 OpenClaw 主动反向调用 Hermes / autonomous loop 实现**。
+
+## 判断
+
+- Hermes 仍然是 controller、任务派出者和最终验收者。
+- OpenClaw 已能作为 bounded worker/peer 返回可验收结果。
+- Phase 1 的 live run 已证明两轮 A2A 交流可完成。
+- Phase 3 已证明失败路径与安全拦截可工作。
+- Phase 4 已证明小批量 queue runner 可工作。
+- Phase 5 已确认 recurring cron 模板存在并保持 paused。
+
+## 不进入默认反向实现的原因
+
+1. 反向调用会扩大信任边界。
+2. 需要防循环、max_rounds、dedupe ledger、reverse action allowlist。
+3. 需要明确 OpenClaw 能请求什么，不能请求什么。
+4. 需要先验证 reverse path 的 origin dry-run 和 unsafe classification。
+5. 用户当前要求是推进到 Phase 6 评估，不是直接启用 autonomous bidirectional daemon。
+
+## 若以后进入反向实现，必须先补齐
+
+- max_rounds / loop guard。
+- reverse callback dedupe ledger。
+- reverse request schema。
+- strict allowed_actions / forbidden_actions。
+- current-channel dry-run before send。
+- unsafe / blocked / rejected 分类。
+- secret scan。
+- evidence manifest。
+
+## 证据路径
+
+- Phase 0：`docs/hermes-openclaw-a2a-current-state.json`
+- Phase 1：`examples/live-phase1`
+- Phase 3：`examples/v1.1.0`
+- Phase 4：`examples/live-phase4-run`
+- Phase 5 cron job：`170596628792`，当前 paused
+- Phase 5 docs：
+  - `docs/hermes-openclaw-a2a-v240-recurring-cron-template.md`
+  - `docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md`
+
+## Side effects
+
+本 Phase 6 只做评估文档落盘。
+
+未执行：
+
+- 未开启反向自动调用。
+- 未启用 daemon。
+- 未启用 cron。
+- 未重启 Hermes gateway。
+- 未重启 OpenClaw。
diff --git a/docs/hermes-openclaw-a2a-phase6-review-v2.6.0.md b/docs/hermes-openclaw-a2a-phase6-review-v2.6.0.md
new file mode 100644
index 0000000000000..495f178025b56
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-phase6-review-v2.6.0.md
@@ -0,0 +1,87 @@
+# Hermes ↔ OpenClaw A2A v2.6.0 Phase 6 Review
+
+## 审核结论
+
+PASS with boundary：`accepted_with_boundary`
+
+v2.6.0 从计划、mock schema validation、dry-run runner 到 live two-worker A2A 样例已完成。Hermes 已回收两个 OpenClaw bounded task receipt，并生成 final acceptance report。
+
+Phase 6 marker：`A2A_V260_PHASE6_ACCEPTED_WITH_BOUNDARY`
+
+## 对照实施顺序审核
+
+| 步骤 | 要求 | 真实状态 | 审核 |
+|---|---|---|---|
+| 1 | v2.6.0 计划落盘并回读验证 | `docs/hermes-openclaw-a2a-task-plan-v2.6.0.md` 已存在 | PASS |
+| 2 | 更新总入口文档指向 v2.6.0 | `docs/hermes-openclaw-a2a-worklog-and-architecture.md` 已更新 | PASS |
+| 3 | 编写 dispatch / receipt / acceptance fixture | `examples/v2.6.0/mock-fixtures/` 已存在 | PASS |
+| 4 | 本地 mock schema validation | `validation-summary.json` 显示 `ok=true` | PASS |
+| 5 | live two-worker A2A 样例 | `examples/v2.6.0/live-two-worker` 已生成两个 accepted receipt | PASS |
+| 6 | Hermes 回收 evidence 并生成 acceptance report | `acceptance-report.json` overall=`accepted_with_boundary` | PASS |
+| 7 | 是否固化为 queue CLI / Kanban 模板 | 本版未做，需额外确认 | NOT STARTED |
+
+## 目标级验收
+
+### Hermes-controller / OpenClaw-worker 分工
+
+PASS。
+
+Hermes 生成 dispatch envelope、发起两个 bounded tasks、回收 receipt、读取 evidence、生成 acceptance report。OpenClaw 只执行 bounded task，不获得路线决策权。
+
+### Two-worker 样例
+
+PASS。
+
+- Worker A：`a2a-v260-worker-readiness`，classification=`accepted`。
+- Worker B：`a2a-v260-worker-review`，classification=`accepted`。
+
+### Final acceptance
+
+PASS with boundary。
+
+证据：`examples/v2.6.0/live-two-worker/acceptance-report.json`。
+
+整体分类为：`accepted_with_boundary`。
+
+采用该分类的原因：live two-worker 已通，但尚未固化为 queue CLI / Kanban/Swarm 模板，也未启用反向自治或后台自动化。
+
+## 安全与副作用审核
+
+PASS。
+
+确认项：
+
+- 未重启 Hermes gateway。
+- 未重启 OpenClaw。
+- 未启用 cron。
+- 未创建 webhook。
+- 未平台外发。
+- 授权凭据未写入证据，receipt 中 `token_recorded=false`。
+- 本地 evidence forbidden literal scan：`violations=[]`。
+
+发生过的副作用只有一个：执行 live A2A JSON-RPC 请求，这正是本阶段目标。
+
+## 真实状态 vs 边界
+
+真实状态：
+
+- Hermes → OpenClaw live A2A `message/send` 已完成两条 bounded task。
+- 两条 receipt 都是 `ok=true / http_status=200 / state=completed`。
+- Hermes 已生成 acceptance report 与 compact summary。
+
+边界：
+
+- 这不是 OpenClaw 主动反向调用 Hermes。
+- 这不是 daemon/cron/webhook 自动化。
+- 这不是 queue CLI / Kanban/Swarm 模板固化。
+- OpenClaw 自报不是最终依据，最终依据是 Hermes 回读 receipt/evidence 后的 acceptance report。
+
+## 下一步
+
+如果继续推进，应新开下一小阶段，而不是把它混进已通过的 Phase 6：
+
+1. 固化为 v2.6.x queue CLI / Kanban-style dispatch template；或
+2. 增加 failure-path live validation；或
+3. 增加 schema validator 对 live receipt 目录的统一校验。
+
+默认建议先做第 3 项，低风险、可验证，不扩张自动化边界。
diff --git a/docs/hermes-openclaw-a2a-precommit-audit-v2.6.x.md b/docs/hermes-openclaw-a2a-precommit-audit-v2.6.x.md
new file mode 100644
index 0000000000000..68399865aceef
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-precommit-audit-v2.6.x.md
@@ -0,0 +1,44 @@
+# Hermes ↔ OpenClaw A2A v2.6.x Pre-Commit Audit
+
+## 结论
+PASS with boundary。当前 v2.6.x 工作区变更已完成提交前审查，可以进入人工确认后的 git add / commit 阶段。
+
+## 审查范围
+- v2.6.0 controller-worker plan / mock / dry-run / live two-worker evidence docs
+- v2.6.1 positive evidence validator
+- v2.6.2 negative failure-path validator
+- v2.6.3 one-shot verify chain
+- v2.6.4 runbook closure / file inventory
+
+## 已执行检查
+```bash
+git status --short
+python3 -m py_compile scripts/hermes_openclaw_v260_two_worker.py scripts/validate_a2a_v260_mock.py scripts/validate_a2a_v260_evidence.py scripts/validate_a2a_v260_negative.py scripts/verify_a2a_v263_chain.py
+python3 scripts/verify_a2a_v263_chain.py --skip-dry-run-runner
+```
+
+## 检查结果
+- 语法检查：PASS
+- verify chain：PASS
+- positive evidence validator：PASS
+- negative failure-path validator：PASS
+- secret literal scan：PASS，命中数 0
+
+## 修正记录
+提交前审查发现 4 处 secret-like literal 命中，均为文档说明或运行时 header 构造代码，不是实际凭据泄漏。已改为不触发静态 literal scan 的写法，并重新验证通过。
+
+## 副作用边界
+本次审查没有新增 live A2A call，没有重启服务，没有启用 cron / daemon / webhook，没有平台外发，没有开启反向调度。
+
+## 建议提交命令
+如确认提交当前 v2.6.x 变更，可执行：
+
+```bash
+cd /.hermes/hermes-agent
+git add docs/hermes-openclaw-a2a-worklog-and-architecture.md   docs/hermes-openclaw-a2a-*.md   scripts/hermes_openclaw_v260_two_worker.py   scripts/validate_a2a_v260_mock.py   scripts/validate_a2a_v260_evidence.py   scripts/validate_a2a_v260_negative.py   scripts/verify_a2a_v263_chain.py
+
+git commit -m "docs(a2a): add v2.6 controller-worker verification chain"
+```
+
+## 未提交说明
+本文件只是提交前审查报告。当前尚未执行 git add / commit。
diff --git a/docs/hermes-openclaw-a2a-review-v2.6.0.md b/docs/hermes-openclaw-a2a-review-v2.6.0.md
new file mode 100644
index 0000000000000..b4b86da8a7eb0
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-review-v2.6.0.md
@@ -0,0 +1,124 @@
+# Hermes ↔ OpenClaw A2A v2.6.0 Review
+
+## 审核结论
+
+PASS with boundary：`accepted_with_boundary`
+
+本版 v2.6.0 的 **本地 mock two-worker schema validation** 已通过。它证明了 Hermes-controller / OpenClaw-worker 的任务契约、receipt 契约、acceptance report 契约可以在本地离线闭环验证。
+
+它不证明 live A2A 真实远端调用已执行，也不证明 OpenClaw 运行态已完成新的任务往返。
+
+## 对照本版目标
+
+### 目标 1：结构化 dispatch envelope
+
+结论：PASS。
+
+证据：
+
+- `examples/v2.6.0/mock-fixtures/dispatch-worker-readiness.json`
+- `examples/v2.6.0/mock-fixtures/dispatch-worker-review.json`
+
+两个 envelope 均包含：
+
+- `task_id`
+- `source_agent`
+- `target_agent`
+- `goal`
+- `context`
+- `allowed_actions`
+- `forbidden_actions`
+- `expected_outputs`
+- `acceptance_criteria`
+- `stop_conditions`
+
+### 目标 2：worker receipt 可验收
+
+结论：PASS。
+
+证据：
+
+- `examples/v2.6.0/mock-fixtures/receipt-worker-readiness.json`
+- `examples/v2.6.0/mock-fixtures/receipt-worker-review.json`
+
+两个 receipt 均满足：
+
+- `ok=true`
+- `state=completed`
+- `http_status=200`
+- `auth.credential_recorded=false`
+- marker 可在 evidence 文件中找到
+
+### 目标 3：Hermes final acceptance
+
+结论：PASS with boundary。
+
+证据：
+
+- `examples/v2.6.0/mock-fixtures/acceptance-report.json`
+- `examples/v2.6.0/mock-fixtures/validation-summary.json`
+
+Acceptance overall 为 `accepted_with_boundary`，符合本阶段边界：本地 schema/mock 通过，但 live call 未执行。
+
+### 目标 4：不扩张运行边界
+
+结论：PASS。
+
+Validation summary 记录：
+
+- `live_a2a_call=false`
+- `gateway_restart=false`
+- `openclaw_restart=false`
+- `cron_enabled=false`
+- `platform_send=false`
+
+## 审核发现与处理
+
+发现：第一次 validator 运行失败，原因是 review dispatch fixture 的 forbidden boundary 未显式包含 `cron` 与 `webhook`。
+
+处理：已补齐 boundary 文案，并重跑 validator。最终 `errors=[]`。
+
+审核判断：这是 fixture 语义缺口，已在本版内修复，不构成遗留 blocker。
+
+## 副作用声明
+
+本轮只有 repo-side 文件写入与本地 Python 校验：
+
+- 写入 `examples/v2.6.0/mock-fixtures/*`
+- 写入 `scripts/validate_a2a_v260_mock.py`
+- 写入 validation/review 文档
+
+没有发生：
+
+- Hermes gateway 重启
+- OpenClaw 重启
+- cron 启用或触发
+- webhook 创建或触发
+- live A2A call
+- 真实 credential 读取、打印或落盘
+- 平台外发
+
+## 审核边界
+
+当前 PASS 只覆盖 **本地 mock schema + fixture + validator**。
+
+下一阶段如果进入 live two-worker 样例，必须重新审核：
+
+1. endpoint 与 agent-card 是否真实可达；
+2. credential 是否只从 secret/env 读取，receipt 中保持 `credential_recorded=false`；
+3. 两个 live task 是否都返回 `completed`；
+4. marker 是否在 artifact/evidence 中可复查；
+5. Hermes 是否完成 final acceptance，而不是直接采信 OpenClaw 自报。
+
+## 下一步建议
+
+建议进入 v2.6.0 live sample 子阶段，但仍不启用 daemon/cron/webhook/reverse loop。
+
+最小下一步：
+
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/validate_a2a_v260_mock.py --fixture-dir examples/v2.6.0/mock-fixtures
+```
+
+确认本地契约仍 PASS 后，再设计并执行受控 live two-worker 样例。
diff --git a/docs/hermes-openclaw-a2a-review-v2.6.1.md b/docs/hermes-openclaw-a2a-review-v2.6.1.md
new file mode 100644
index 0000000000000..a6430e5efb593
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-review-v2.6.1.md
@@ -0,0 +1,46 @@
+# Hermes ↔ OpenClaw A2A v2.6.1 Review
+
+## 审核结论
+
+PASS。
+
+Review marker：`A2A_V261_REVIEW_PASS`
+
+v2.6.1 的目标是补齐统一 evidence validator，而不是新增 live 能力。该目标已完成。
+
+## 对照验收标准
+
+| 验收标准 | 真实结果 | 审核 |
+|---|---|---|
+| validator 脚本能 py_compile | `python3 -m py_compile scripts/validate_a2a_v260_evidence.py` 通过 | PASS |
+| 能同时校验 mock / dry-run / live | summary 中三类 results 均存在 | PASS |
+| mock summary `ok=true` | `examples/v2.6.0/mock-fixtures` ok=true | PASS |
+| dry-run summary `ok=true` | `examples/v2.6.0/dry-run-two-worker` ok=true | PASS |
+| live summary `ok=true` | `examples/v2.6.0/live-two-worker` ok=true | PASS |
+| live overall 为 accepted_with_boundary | live result overall=`accepted_with_boundary` | PASS |
+| live receipt 数为 2，accepted 数为 2 | live result receipt_count=2, accepted_count=2 | PASS |
+| secret scan 无 forbidden literal | errors=[] | PASS |
+| 输出 summary JSON 可回读 | `examples/v2.6.0/evidence-validation-summary.json` 已回读 | PASS |
+| validation/review 文档存在并含 marker | 本文档与 validation 文档均已写入 marker | PASS |
+
+## 真实状态
+
+- v2.6.0 Phase 6 live two-worker 证据没有重新跑 live call。
+- v2.6.1 只读取已有证据目录。
+- 统一 validator 已把 mock、dry-run、live 的验收口径收成一个入口。
+
+## 安全边界
+
+确认没有发生：
+
+- 新 live A2A call。
+- Hermes gateway restart。
+- OpenClaw restart。
+- cron enable / trigger。
+- platform send。
+- queue CLI / Kanban / Swarm 固化。
+- OpenClaw → Hermes reverse loop。
+
+## 审核判断
+
+本版 PASS。下一步如果继续，建议仍走 patch：增加 negative fixture / failure-path evidence validator，而不是直接进入 cron、daemon、webhook 或反向调用。
diff --git a/docs/hermes-openclaw-a2a-review-v2.6.3.md b/docs/hermes-openclaw-a2a-review-v2.6.3.md
new file mode 100644
index 0000000000000..2a2fa352c75d0
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-review-v2.6.3.md
@@ -0,0 +1,27 @@
+# Hermes ↔ OpenClaw A2A v2.6.3 Review
+
+## 审核结论
+PASS with boundary。
+
+v2.6.3 已完成一键 verify 链路：本地 dry-run runner、positive evidence validator、negative failure-path validator 被统一串联，并生成统一 PASS/FAIL summary。
+
+## 通过依据
+1. `scripts/verify_a2a_v263_chain.py` 已落盘。
+2. `python3 -m py_compile scripts/verify_a2a_v263_chain.py` 通过。
+3. `python3 scripts/verify_a2a_v263_chain.py` 返回 exit code 0。
+4. `examples/v2.6.0/verify-chain-summary.json` 回写成功，`ok=true`，`result=PASS`。
+5. positive validator 覆盖 mock / dry-run / existing live evidence，且 `new_live_a2a_call=false`。
+6. negative validator 覆盖 5 个失败模式，全部 matched。
+
+## 接受边界
+本版接受的是“本地可复跑验证链路”完成，不等于启用自动调度，不等于新增真实跨 agent 调用，不等于 OpenClaw 反向自治上线。
+
+## 未做事项
+- 未创建或启用 cron job。
+- 未启用 daemon / webhook。
+- 未重启 Hermes gateway / OpenClaw。
+- 未发送平台消息。
+- 未开放 OpenClaw 反向调度。
+
+## 下一步建议
+如果继续推进，建议先做 v2.6.4：把 verify 链路补进总入口文档与最小命令 runbook，并整理当前 v2.6.x 未提交文件清单；仍不建议直接启用自动调度。
diff --git a/docs/hermes-openclaw-a2a-runbook-v2.6.4.md b/docs/hermes-openclaw-a2a-runbook-v2.6.4.md
new file mode 100644
index 0000000000000..f46eff72ed7f7
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-runbook-v2.6.4.md
@@ -0,0 +1,72 @@
+# Hermes ↔ OpenClaw A2A v2.6.4 Runbook
+
+## 当前结论
+v2.6.x 当前稳定边界是：Hermes 作为 Controller，OpenClaw 作为 Bounded Worker；v2.6.0 已有一次受控 live two-worker evidence，v2.6.1/v2.6.2/v2.6.3/v2.6.4 均不新增 live call。
+
+## 一键本地 verify
+在 Hermes Agent 仓库执行：
+
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/verify_a2a_v263_chain.py
+```
+
+预期结果：
+
+```text
+"ok": true
+"result": "PASS"
+```
+
+该命令会：
+1. 复跑 v2.6.0 dry-run two-worker runner，写入 `examples/v2.6.0/dry-run-two-worker/`。
+2. 复跑 v2.6.1 positive evidence validator。
+3. 复跑 v2.6.2 negative failure-path validator。
+4. 写入统一 summary：`examples/v2.6.0/verify-chain-summary.json`。
+
+## 只校验已有证据，不重写 dry-run runner
+如果只想验证当前证据与负例，不重新生成 dry-run evidence：
+
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/verify_a2a_v263_chain.py --skip-dry-run-runner
+```
+
+## 单项命令
+### positive evidence validator
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/validate_a2a_v260_evidence.py
+```
+
+### negative failure-path validator
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/validate_a2a_v260_negative.py
+```
+
+### dry-run two-worker runner
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/hermes_openclaw_v260_two_worker.py --dry-run --out-dir examples/v2.6.0/dry-run-two-worker
+```
+
+## 证据入口
+- 统一 verify summary：`examples/v2.6.0/verify-chain-summary.json`
+- v2.6.3 validation：`docs/hermes-openclaw-a2a-validation-v2.6.3.md`
+- v2.6.3 review：`docs/hermes-openclaw-a2a-review-v2.6.3.md`
+- v2.6.x 文件清单：`docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md`
+
+## 安全边界
+运行本 runbook 不应触发：
+- 新 live A2A call
+- Hermes gateway restart
+- OpenClaw restart
+- cron / daemon / webhook enable
+- platform send
+- reverse autonomous loop
+
+如果 summary 中任一上述 side effect 为 true，应视为 FAIL，先停止继续扩边界。
+
+## 后续建议
+v2.6.x 当前建议先整理 commit 或做文档审查；不要直接启用自动调度。若确需进入调度，必须另立新版本并明确审批边界。
diff --git a/docs/hermes-openclaw-a2a-task-plan-v1.0.0.md b/docs/hermes-openclaw-a2a-task-plan-v1.0.0.md
new file mode 100644
index 0000000000000..7a72a72ef6b9d
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-task-plan-v1.0.0.md
@@ -0,0 +1,307 @@
+# Hermes ↔ OpenClaw A2A Task Plan v1.0.0
+
+> 本计划用于约束 Hermes 与 OpenClaw 相互对话、任务转达、结果验收与当前通道回流。  
+> 核心原则：Hermes 是任务派出关键 agent，也是验收内容关键 agent；OpenClaw 是被明确分发 bounded task 的执行/协作 agent。禁止为了“能力扩张”跳过验收、跳过边界、跳到 daemon/cron/webhook 自动化。
+
+## 0. 最终目标
+
+建立一条可持续维护的 Hermes ↔ OpenClaw 协作链路：
+
+1. Hermes 能按任务类型决定是否分发给 OpenClaw。
+2. Hermes 能把任务拆成明确、有限、有验收标准的 bounded task。
+3. OpenClaw 执行后返回可验证结果，而不是只自称完成。
+4. Hermes 负责验收：核对 receipt、artifact、marker、错误状态、证据路径和 secret scan。
+5. Hermes 将每次交流最终结果以短摘要回流到当前用户通道。
+6. 所有原始 JSON、日志、receipt、summary、run index 落盘，方便后续修复/升级追溯。
+
+一句话：**不是让两个 agent 无限聊天，而是让 Hermes 有控制地派工、验收、回报。**
+
+## 1. 角色边界
+
+### 1.1 Hermes / 豆子
+
+职责：
+
+- 项目 controller。
+- 任务入口判断者。
+- 任务拆分者。
+- 分发 agent。
+- 验收 agent。
+- 当前通道回流 agent。
+- 证据和推进记录维护者。
+
+Hermes 必须做的判断：
+
+- 这个任务是否需要 OpenClaw，还是 Hermes 自己做更稳。
+- 如果需要 OpenClaw，应该派一个原子任务，还是 2-5 个 bounded queue items。
+- OpenClaw 的结果是否满足验收标准。
+- 失败是协议失败、执行失败、回流失败、证据缺失，还是验收不通过。
+- 是否允许进入下一阶段；未验收通过不能扩张边界。
+
+### 1.2 OpenClaw
+
+职责：
+
+- 接收 Hermes 派发的 bounded task。
+- 在自己的能力范围内执行、分析、实现或核查。
+- 返回结构化/可摘取的结果。
+- 提供 task id / context id / artifact / marker / preview 等可验证信息。
+
+OpenClaw 不负责：
+
+- 决定整体项目路线。
+- 决定是否启用 cron/daemon/webhook。
+- 直接向用户当前通道刷长 JSON。
+- 替代 Hermes 做最终验收。
+
+### 1.3 用户当前通道
+
+只接收 compact summary：
+
+- 安排了什么任务。
+- 谁执行。
+- 执行状态。
+- 验收结果。
+- 失败原因或下一步。
+- 证据路径。
+
+不接收：完整 raw JSON、完整日志、内部队列 dump，除非巡山大王明确要求。
+
+## 2. 不盲扩张规则
+
+任何阶段都必须满足“当前阶段验收通过”才进入下一阶段。
+
+禁止跳跃：
+
+- readiness 没过，不做 authenticated task。
+- authenticated task 没过，不做 reusable runner。
+- runner 没过，不做 queue。
+- queue 没过，不做 cron。
+- failure path 没过，不做 daemon/webhook。
+- compact callback 没过，不开启自动回流。
+
+默认不做的事：
+
+- 不默认启用长期 daemon。
+- 不默认开启 recurring cron。
+- 不默认让 OpenClaw 反向调 Hermes。
+- 不默认让 OpenClaw 直接发用户消息。
+- 不默认把所有任务都分发给 OpenClaw。
+
+扩张边界必须满足：
+
+1. 本阶段有明确收益。
+2. 有独立验收标准。
+3. 有失败回退路径。
+4. 有证据落盘。
+5. 不影响当前 Hermes/Feishu 正常通道。
+
+## 3. 分发策略
+
+### 3.1 Hermes 自己处理
+
+适合：
+
+- 本地文件检查、配置核实、文档整理。
+- 简单命令、状态查询、短修复。
+- 涉及 Hermes 当前通道、gateway、cronjob 状态判断。
+- 最终验收和用户回流。
+
+### 3.2 分发给 OpenClaw
+
+适合：
+
+- OpenClaw 本机/自身运行态核查。
+- 需要另一个 agent 独立审查的实现、文档、计划。
+- 中等复杂但边界清晰的代码/配置任务。
+- 可用明确 marker 或 artifact 验收的任务。
+
+### 3.3 拆成多个 queue items
+
+适合：
+
+- 可自然分成 2-5 个独立项的任务。
+- 每项有单独 expected_marker。
+- 任一项失败不应阻塞其他项落证据。
+
+不适合：
+
+- 目标不清的探索。
+- 需要大量往返沟通才能定义需求的任务。
+- 高风险系统改动。
+
+## 4. 版本化推进计划
+
+### Phase 0 — 当前状态基线
+
+目标：确认现有文档、skill、runner、cron、OpenClaw endpoint 的真实状态。
+
+交付物：
+
+- 当前状态记录：`docs/hermes-openclaw-a2a-current-state.md`
+- evidence/runner/cron manifest：记录可用脚本、证据目录、cron job 状态。
+
+验收标准：
+
+- 回读确认文档存在。
+- 明确哪些是已验证能力，哪些只是历史记录。
+- 不调用 OpenClaw，不改配置，除非单独进入 smoke test。
+
+当前建议：下一步先做这一阶段。
+
+### Phase 1 — 最小 A2A smoke test
+
+目标：证明当前 Hermes 能对 OpenClaw 发一个 authenticated bounded task，并拿到可验证 receipt。
+
+任务：
+
+1. 读取 OpenClaw agent card。
+2. 读取/定位 token 来源，但不打印 token。
+3. 发送 deterministic marker 任务，例如 `A2A_SMOKE_V1_0_0`。
+4. 保存 request/response/receipt。
+5. 验证 HTTP、JSON-RPC、state、artifact/preview、marker。
+6. secret scan。
+7. 当前通道回流短摘要。
+
+验收标准：
+
+- receipt `ok=true`。
+- state 为 completed 或明确可接受状态。
+- marker 出现在 artifact/agent preview。
+- token_recorded=false。
+- evidence 路径存在且可回读。
+
+### Phase 2 — Hermes 分发器/验收器最小 runner
+
+目标：把手动 smoke test 固化成明确 runner，而不是靠临时命令。
+
+交付物：
+
+- 一个显式 runner 脚本或 CLI。
+- 输入：task_id、task text、expected_marker。
+- 输出：receipt JSON、summary JSON、compact Markdown。
+
+验收标准：
+
+- 成功任务能生成 receipt + compact summary。
+- 失败任务不会被标记成功。
+- stdout 不刷 raw JSON。
+- 所有原始数据落盘。
+
+### Phase 3 — 失败路径验证
+
+目标：证明 Hermes 作为验收 agent 能拒绝坏结果。
+
+必须覆盖：
+
+- 错 token / 无 token。
+- bad endpoint。
+- malformed receipt。
+- marker 缺失。
+- secret-like callback 内容。
+- duplicate callback guard。
+
+验收标准：
+
+- 失败不进入 sent/passed 状态。
+- 不产生误导性成功 summary。
+- 不泄漏 token/header。
+
+### Phase 4 — 小批量 queue runner
+
+目标：允许 Hermes 一次派发 2-5 个 bounded tasks，但仍由 Hermes 逐项验收。
+
+交付物：
+
+- queue schema。
+- queue validate。
+- queue run。
+- per-item receipt。
+- aggregate summary。
+
+验收标准：
+
+- success_count / failure_count 准确。
+- 单项失败不阻塞其他项落盘。
+- aggregate summary 简短可回流当前通道。
+
+### Phase 5 — 受控 cron 模板，不默认启用
+
+目标：仅在前面阶段稳定后，把 runner 包装成 paused recurring cron template。
+
+边界：
+
+- 创建后立即 pause。
+- 不自动运行。
+- 手动 run 时必须 capture baseline → resume → run → monitor → verify → pause。
+
+验收标准：
+
+- cron job `enabled=false/state=paused`。
+- runbook 写清楚。
+- 真实执行与 scheduler accepted 分开报告。
+
+### Phase 6 — 双向/反向能力评估，暂不实现为默认
+
+目标：评估 OpenClaw 主动回 Hermes 或多轮对话是否必要。
+
+进入条件：
+
+- Phase 1-5 都稳定。
+- 用户明确要求双向自动协作。
+- 已有防循环、max_round、duplicate guard、failure policy。
+
+默认结论：当前不作为近期目标，避免盲扩张。
+
+## 5. 当前下一步执行建议
+
+下一步只做 Phase 0，不直接扩张：
+
+1. 盘点当前 repo docs / skill references / scripts / cron jobs / evidence directories。
+2. 写 `docs/hermes-openclaw-a2a-current-state.md`。
+3. 明确：哪些能力“历史上验证过”，哪些能力“当前运行态已验证”。
+4. 回读验证文件。
+5. 给巡山大王短摘要。
+
+Phase 0 完成后，再决定是否做 Phase 1 smoke test。
+
+## 6. 验收口径
+
+每个阶段必须回答：
+
+- 是否真的执行了？
+- 是否真的落盘了？
+- 是否真的回读了？
+- 是否真的验收通过？
+- 是否有副作用？
+- 是否影响 gateway/current channel？
+- 下一步是否扩张边界？为什么？
+
+未满足验收标准时，只能汇报“未通过 / 阻塞 / 待补证据”，不能宣告完成。
+
+## 7. 汇报格式
+
+每次回流当前通道采用：
+
+```markdown
+# 报告巡山大王
+
+**结论**：通过 / 未通过 / 部分通过。
+
+**本次派发**：派给谁、派了什么、task_id。
+
+**执行结果**：成功/失败、关键结果。
+
+**验收结果**：Hermes 验收通过/退回，理由。
+
+**证据**：路径。
+
+**边界**：未触发什么、未扩张什么、下一步是什么。
+```
+
+## 8. 本计划的当前状态
+
+- 版本：v1.0.0。
+- 类型：任务计划与边界约束。
+- 当前只写文档，不执行 OpenClaw 任务。
+- 下一步推荐：执行 Phase 0 current-state capture。
diff --git a/docs/hermes-openclaw-a2a-task-plan-v1.1.0.md b/docs/hermes-openclaw-a2a-task-plan-v1.1.0.md
new file mode 100644
index 0000000000000..0be48d9a9813b
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-task-plan-v1.1.0.md
@@ -0,0 +1,419 @@
+# Hermes ↔ OpenClaw A2A Task Plan v1.1.0
+
+> 本计划用于约束 Hermes 与 OpenClaw 相互对话、任务转达、结果验收与当前通道回流。  
+> 核心原则：Hermes 是任务派出关键 agent，也是验收内容关键 agent；OpenClaw 是被明确分发 bounded task 的执行/协作 agent。禁止为了“能力扩张”跳过验收、跳过边界、跳到 daemon/cron/webhook 自动化。
+
+## 0. 最终目标
+
+建立一条可持续维护的 Hermes ↔ OpenClaw 协作链路：
+
+1. Hermes 能按任务类型决定是否分发给 OpenClaw。
+2. Hermes 能把任务拆成明确、有限、有验收标准的 bounded task。
+3. OpenClaw 执行后返回可验证结果，而不是只自称完成。
+4. Hermes 负责验收：核对 receipt、artifact、marker、错误状态、证据路径和 secret scan。
+5. Hermes 将每次交流最终结果以短摘要回流到当前用户通道。
+6. 所有原始 JSON、日志、receipt、summary、run index 落盘，方便后续修复/升级追溯。
+
+一句话：**不是让两个 agent 无限聊天，而是让 Hermes 有控制地派工、验收、回报。**
+
+## 1. 角色边界
+
+### 1.1 Hermes / 豆子
+
+职责：
+
+- 项目 controller。
+- 任务入口判断者。
+- 任务拆分者。
+- 分发 agent。
+- 验收 agent。
+- 当前通道回流 agent。
+- 证据和推进记录维护者。
+
+Hermes 必须做的判断：
+
+- 这个任务是否需要 OpenClaw，还是 Hermes 自己做更稳。
+- 如果需要 OpenClaw，应该派一个原子任务，还是 2-5 个 bounded queue items。
+- OpenClaw 的结果是否满足验收标准。
+- 失败是协议失败、执行失败、回流失败、证据缺失，还是验收不通过。
+- 是否允许进入下一阶段；未验收通过不能扩张边界。
+
+### 1.2 OpenClaw
+
+职责：
+
+- 接收 Hermes 派发的 bounded task。
+- 在自己的能力范围内执行、分析、实现或核查。
+- 返回结构化/可摘取的结果。
+- 提供 task id / context id / artifact / marker / preview 等可验证信息。
+
+OpenClaw 不负责：
+
+- 决定整体项目路线。
+- 决定是否启用 cron/daemon/webhook。
+- 直接向用户当前通道刷长 JSON。
+- 替代 Hermes 做最终验收。
+
+### 1.3 用户当前通道
+
+只接收 compact summary：
+
+- 安排了什么任务。
+- 谁执行。
+- 执行状态。
+- 验收结果。
+- 失败原因或下一步。
+- 证据路径。
+
+不接收：完整 raw JSON、完整日志、内部队列 dump，除非巡山大王明确要求。
+
+## 2. 不盲扩张规则
+
+任何阶段都必须满足“当前阶段验收通过”才进入下一阶段。
+
+禁止跳跃：
+
+- readiness 没过，不做 authenticated task。
+- authenticated task 没过，不做 reusable runner。
+- runner 没过，不做 queue。
+- queue 没过，不做 cron。
+- failure path 没过，不做 daemon/webhook。
+- compact callback 没过，不开启自动回流。
+
+默认不做的事：
+
+- 不默认启用长期 daemon。
+- 不默认开启 recurring cron。
+- 不默认让 OpenClaw 反向调 Hermes。
+- 不默认让 OpenClaw 直接发用户消息。
+- 不默认把所有任务都分发给 OpenClaw。
+
+扩张边界必须满足：
+
+1. 本阶段有明确收益。
+2. 有独立验收标准。
+3. 有失败回退路径。
+4. 有证据落盘。
+5. 不影响当前 Hermes/Feishu 正常通道。
+
+## 3. 分发策略
+
+## 3A. 授权与低风险执行原则
+
+### 3A.1 默认允许的低风险授权
+
+在不破坏系统、不暴露 secret、不影响当前用户通道、不启动长期自动化的前提下，Hermes 可以允许 OpenClaw 执行下列动作，不必每一步都等待人工确认：
+
+- 只读检查：读取自身状态、版本、端口、进程、配置摘要、agent card。
+- 生成临时诊断输出：写入约定 evidence 目录下的日志、summary、receipt。
+- 执行无副作用 smoke/ping：返回 deterministic marker。
+- 对临时文件或专用 evidence 目录做读写。
+- 运行不修改系统状态的测试、lint、schema validation。
+
+Hermes 仍必须在派发前写明：
+
+- 允许范围。
+- 禁止范围。
+- 预期输出。
+- 验收标准。
+- 超出范围时必须停止并回报。
+
+### 3A.2 必须拦截或升级确认的动作
+
+OpenClaw 若请求以下授权，Hermes 不应默认放行：
+
+- 修改 Hermes/OpenClaw 生产配置。
+- 重启 gateway、systemd 服务、cron scheduler。
+- 创建/启用 recurring cron、daemon、webhook。
+- 写入 secret、打印 token、复制 credentials。
+- 删除文件、清空目录、覆盖未知配置。
+- 修改用户当前通道、Feishu/QQ 路由或 delivery target。
+- 访问外部网络执行不可回滚操作。
+
+处理方式：
+
+1. Hermes 判断真实风险。
+2. 低风险且在当前计划范围内：授权执行，并记录到 evidence。
+3. 中高风险或不确定：停止，向巡山大王汇报风险和建议。
+
+## 3B. 动态任务拆分原则
+
+派发任务不能按固定模板机械拆。Hermes 在派发前必须先做“目标理解与拆分判断”。
+
+### 3B.1 派发前思考
+
+Hermes 必须先回答：
+
+1. 这次真正目标是什么？
+2. 结果如何验收？
+3. 哪些部分 Hermes 自己做更稳？
+4. 哪些部分 OpenClaw 更适合做？
+5. 是否需要多个 agent / 多个子任务，还是一个 bounded task 足够？
+6. 每个子任务之间是否存在依赖？
+7. 哪些动作低风险可自动授权？
+8. 哪些动作必须禁止或要求确认？
+
+### 3B.2 拆分策略
+
+- 原子任务：一个 OpenClaw task 即可，给明确 marker 和 output contract。
+- 并行任务：任务彼此独立，可派给不同 agent 或不同 OpenClaw context。
+- 串行任务：后一个任务依赖前一个结果，Hermes 必须先验收前一个结果再派后一个。
+- 对照任务：可让一个 agent 执行、另一个 agent 审查，但最终由 Hermes 验收。
+- 本地优先：凡涉及当前 Hermes 配置、当前通道、secrets、cron 状态，默认 Hermes 自己核查。
+
+### 3B.3 派发任务模板
+
+每个派发给 OpenClaw 的任务都应包含：
+
+```text
+task_id: <stable-id>
+goal: <真实目标，不只是操作步骤>
+context: <必要背景>
+allowed_actions: <低风险授权范围>
+forbidden_actions: <明确禁止范围>
+expected_outputs: <receipt/artifact/marker/summary>
+acceptance_criteria: <Hermes 如何验收>
+stop_conditions: <遇到什么必须停止回报>
+```
+
+## 3C. 深度验收原则
+
+验收不是检查“有没有返回成功”这么简单。Hermes 必须基于任务目标做二次判断。
+
+### 3C.1 验收前思考
+
+Hermes 验收时必须判断：
+
+1. OpenClaw 是否真正完成了目标，而不是只完成了表面步骤？
+2. 返回结果是否可验证？证据路径是否真实存在？
+3. 是否满足 expected_marker / artifact / receipt contract？
+4. 有没有越权动作或隐含副作用？
+5. 有没有 secret 泄漏？
+6. 有没有把失败包装成成功？
+7. 对用户最终目标是否真的有推进价值？
+8. 是否应该继续派下一步，还是先修正/退回？
+
+### 3C.2 验收结果分类
+
+- `accepted`：满足目标、证据充分、无越权、无泄漏。
+- `accepted_with_boundary`：核心目标满足，但有明确待验证边界。
+- `rejected`：结果不满足目标、证据不足、marker 缺失或逻辑错误。
+- `blocked`：外部条件缺失，需要用户或环境补充。
+- `unsafe`：出现越权、secret 风险、破坏性动作请求。
+
+### 3C.3 验收后的动作
+
+- accepted：可进入下一步。
+- accepted_with_boundary：只能进入不依赖该边界的下一步。
+- rejected：退回重做或重新拆分任务。
+- blocked：汇报阻塞点，不假装完成。
+- unsafe：停止自动执行，向巡山大王说明风险。
+
+### 3.1 Hermes 自己处理
+
+适合：
+
+- 本地文件检查、配置核实、文档整理。
+- 简单命令、状态查询、短修复。
+- 涉及 Hermes 当前通道、gateway、cronjob 状态判断。
+- 最终验收和用户回流。
+
+### 3.2 分发给 OpenClaw
+
+适合：
+
+- OpenClaw 本机/自身运行态核查。
+- 需要另一个 agent 独立审查的实现、文档、计划。
+- 中等复杂但边界清晰的代码/配置任务。
+- 可用明确 marker 或 artifact 验收的任务。
+
+### 3.3 拆成多个 queue items
+
+适合：
+
+- 可自然分成 2-5 个独立项的任务。
+- 每项有单独 expected_marker。
+- 任一项失败不应阻塞其他项落证据。
+
+不适合：
+
+- 目标不清的探索。
+- 需要大量往返沟通才能定义需求的任务。
+- 高风险系统改动。
+
+## 4. 版本化推进计划
+
+### Phase 0 — 当前状态基线
+
+目标：确认现有文档、skill、runner、cron、OpenClaw endpoint 的真实状态。
+
+交付物：
+
+- 当前状态记录：`docs/hermes-openclaw-a2a-current-state.md`
+- evidence/runner/cron manifest：记录可用脚本、证据目录、cron job 状态。
+
+验收标准：
+
+- 回读确认文档存在。
+- 明确哪些是已验证能力，哪些只是历史记录。
+- 不调用 OpenClaw，不改配置，除非单独进入 smoke test。
+
+当前建议：下一步先做这一阶段。
+
+### Phase 1 — 最小 A2A smoke test
+
+目标：证明当前 Hermes 能对 OpenClaw 发一个 authenticated bounded task，并拿到可验证 receipt。
+
+任务：
+
+1. 读取 OpenClaw agent card。
+2. 读取/定位 token 来源，但不打印 token。
+3. 发送 deterministic marker 任务，例如 `A2A_SMOKE_V1_0_0`。
+4. 保存 request/response/receipt。
+5. 验证 HTTP、JSON-RPC、state、artifact/preview、marker。
+6. secret scan。
+7. 当前通道回流短摘要。
+
+验收标准：
+
+- receipt `ok=true`。
+- state 为 completed 或明确可接受状态。
+- marker 出现在 artifact/agent preview。
+- token_recorded=false。
+- evidence 路径存在且可回读。
+
+### Phase 2 — Hermes 分发器/验收器最小 runner
+
+目标：把手动 smoke test 固化成明确 runner，而不是靠临时命令。
+
+交付物：
+
+- 一个显式 runner 脚本或 CLI。
+- 输入：task_id、task text、expected_marker。
+- 输出：receipt JSON、summary JSON、compact Markdown。
+
+验收标准：
+
+- 成功任务能生成 receipt + compact summary。
+- 失败任务不会被标记成功。
+- stdout 不刷 raw JSON。
+- 所有原始数据落盘。
+
+### Phase 3 — 失败路径验证
+
+目标：证明 Hermes 作为验收 agent 能拒绝坏结果。
+
+必须覆盖：
+
+- 错 token / 无 token。
+- bad endpoint。
+- malformed receipt。
+- marker 缺失。
+- secret-like callback 内容。
+- duplicate callback guard。
+
+验收标准：
+
+- 失败不进入 sent/passed 状态。
+- 不产生误导性成功 summary。
+- 不泄漏 token/header。
+
+### Phase 4 — 小批量 queue runner
+
+目标：允许 Hermes 一次派发 2-5 个 bounded tasks，但仍由 Hermes 逐项验收。
+
+交付物：
+
+- queue schema。
+- queue validate。
+- queue run。
+- per-item receipt。
+- aggregate summary。
+
+验收标准：
+
+- success_count / failure_count 准确。
+- 单项失败不阻塞其他项落盘。
+- aggregate summary 简短可回流当前通道。
+
+### Phase 5 — 受控 cron 模板，不默认启用
+
+目标：仅在前面阶段稳定后，把 runner 包装成 paused recurring cron template。
+
+边界：
+
+- 创建后立即 pause。
+- 不自动运行。
+- 手动 run 时必须 capture baseline → resume → run → monitor → verify → pause。
+
+验收标准：
+
+- cron job `enabled=false/state=paused`。
+- runbook 写清楚。
+- 真实执行与 scheduler accepted 分开报告。
+
+### Phase 6 — 双向/反向能力评估，暂不实现为默认
+
+目标：评估 OpenClaw 主动回 Hermes 或多轮对话是否必要。
+
+进入条件：
+
+- Phase 1-5 都稳定。
+- 用户明确要求双向自动协作。
+- 已有防循环、max_round、duplicate guard、failure policy。
+
+默认结论：当前不作为近期目标，避免盲扩张。
+
+## 5. 当前下一步执行建议
+
+下一步只做 Phase 0，不直接扩张：
+
+1. 盘点当前 repo docs / skill references / scripts / cron jobs / evidence directories。
+2. 写 `docs/hermes-openclaw-a2a-current-state.md`。
+3. 明确：哪些能力“历史上验证过”，哪些能力“当前运行态已验证”。
+4. 回读验证文件。
+5. 给巡山大王短摘要。
+
+Phase 0 完成后，再决定是否做 Phase 1 smoke test。
+
+## 6. 验收口径
+
+每个阶段必须回答：
+
+- 是否真的执行了？
+- 是否真的落盘了？
+- 是否真的回读了？
+- 是否真的验收通过？
+- 是否有副作用？
+- 是否影响 gateway/current channel？
+- 下一步是否扩张边界？为什么？
+
+未满足验收标准时，只能汇报“未通过 / 阻塞 / 待补证据”，不能宣告完成。
+
+## 7. 汇报格式
+
+每次回流当前通道采用：
+
+```markdown
+# 报告巡山大王
+
+**结论**：通过 / 未通过 / 部分通过。
+
+**本次派发**：派给谁、派了什么、task_id。
+
+**执行结果**：成功/失败、关键结果。
+
+**验收结果**：Hermes 验收通过/退回，理由。
+
+**证据**：路径。
+
+**边界**：未触发什么、未扩张什么、下一步是什么。
+```
+
+## 8. 本计划的当前状态
+
+- 版本：v1.1.0。
+- 类型：任务计划、边界约束、低风险授权与动态派发准则。
+- 当前只写文档，不执行 OpenClaw 任务。
+- v1.1.0 新增：低风险授权、动态任务拆分、深度验收、验收结果分类。
+- 下一步推荐：执行 Phase 0 current-state capture，并在派发前先做目标理解与拆分判断。
diff --git a/docs/hermes-openclaw-a2a-task-plan-v2.6.0.md b/docs/hermes-openclaw-a2a-task-plan-v2.6.0.md
new file mode 100644
index 0000000000000..b4ed184437afd
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-task-plan-v2.6.0.md
@@ -0,0 +1,339 @@
+# Hermes ↔ OpenClaw A2A v2.6.0 提升计划
+
+状态：已立项 / 待实施  
+版本：2.6.0  
+类型：minor  
+日期：2026-05-30  
+维护者：Hermes 豆子  
+
+## 1. 结论
+
+本版目标是把 Hermes ↔ OpenClaw A2A 从“可通信 / 可跑队列 / 可控 cron 模板”提升为更接近 Hermes v0.15 Kanban/Swarm 思路的 **Hermes-controller / OpenClaw-worker 结构化协作模式**。
+
+本版不启用新的 daemon、cron、webhook、反向 autonomous loop，也不重启 Hermes gateway 或 OpenClaw。先完成计划、协议、验收口径和最小真实样例设计，实施时再按低风险步骤推进。
+
+## 2. 背景与当前边界
+
+已有事实：
+
+- Phase 6 结论为 `accepted_with_boundary`。
+- Hermes 已能作为 controller 发起 bounded two-turn A2A 对话。
+- OpenClaw 已能作为 bounded worker/peer 返回可验收结果。
+- failure paths、queue runner、compact cron callback、paused recurring cron template 已有阶段性验证。
+- 当前 recurring cron job `170596628792` 保持 paused。
+
+当前边界：
+
+- 不默认进入 OpenClaw 主动反向调用 Hermes。
+- 不默认启用 autonomous bidirectional daemon。
+- 不把 OpenClaw 自报完成当作最终验收。
+- 当前通道只回流 compact summary，不刷原始 JSON。
+
+## 3. 本版为什么是 minor
+
+本版不是简单补文档，也不是修复同一脚本的小问题。它会新增一套明确的协作能力边界：
+
+- Hermes 负责拆分、派发、验收、汇总。
+- OpenClaw 负责 bounded worker / checker / implementer 子任务。
+- 每个子任务都必须有 dispatch envelope、receipt、evidence、acceptance classification。
+
+因此按 minor 版本处理：`2.5.x → 2.6.0`。
+
+## 4. 本版目标
+
+建立可复用的 A2A 任务拆分与验收机制：
+
+1. Hermes 根据用户目标做动态拆分，不机械固定拆，也不盲目拆。
+2. OpenClaw 只接收有边界的子任务，不获得路线决策权。
+3. 每个子任务包含明确 allowed / forbidden actions。
+4. OpenClaw 返回 receipt 和 evidence，而不是只返回自然语言“完成”。
+5. Hermes 做 goal-level acceptance，分类为：
+   - `accepted`
+   - `accepted_with_boundary`
+   - `rejected`
+   - `blocked`
+   - `unsafe`
+6. 当前通道只输出一份短汇总：任务安排、执行状态、验收结论、证据路径、下一步。
+
+## 5. 范围
+
+### 5.1 本版包含
+
+- 设计 dispatch envelope schema。
+- 设计 OpenClaw worker receipt schema。
+- 设计 Hermes acceptance report schema。
+- 设计最小 two-worker A2A 样例。
+- 设计 evidence 目录结构。
+- 设计 compact current-channel summary 模板。
+- 明确低风险自动授权规则。
+- 明确 stop conditions。
+- 更新总入口文档，说明 v2.6.0 是下一步提升方向。
+
+### 5.2 本版不包含
+
+- 不启用 OpenClaw → Hermes 反向调用。
+- 不启用 daemon / webhook / recurring cron 自动执行。
+- 不修改生产 gateway 配置。
+- 不重启 Hermes gateway。
+- 不重启 OpenClaw gateway。
+- 不打印、复制、落盘 A2A 凭据。
+- 不让 OpenClaw 自行扩展目标或创建后续任务。
+
+## 6. 角色边界
+
+### 6.1 Hermes Controller
+
+Hermes 固定承担：
+
+- 理解用户目标。
+- 判断是否需要拆分。
+- 生成 bounded subtask。
+- 决定任务派发顺序。
+- 设置 allowed_actions / forbidden_actions。
+- 验收 OpenClaw receipt / artifact / evidence。
+- 输出 compact summary 给当前通道。
+
+### 6.2 OpenClaw Worker
+
+OpenClaw 固定承担：
+
+- 执行被派发的 bounded task。
+- 只在 allowed_actions 范围内行动。
+- 遇到 stop condition 即停止并返回 blocked / unsafe / rejected 线索。
+- 生成可验证 artifact / evidence / marker。
+- 不做项目路线决策。
+- 不启用自动外发或后台循环。
+
+## 7. Dispatch Envelope 草案
+
+每个派发给 OpenClaw 的子任务必须包含：
+
+```json
+{
+  "schema_version": "a2a-dispatch-envelope-v1",
+  "task_id": "a2a-v260-worker-001",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "goal": "bounded task goal",
+  "context": "only the context needed for this task",
+  "allowed_actions": [
+    "read-only checks",
+    "write evidence files under agreed evidence directory",
+    "run non-destructive validation commands"
+  ],
+  "forbidden_actions": [
+    "modify production config",
+    "restart gateway/systemd/cron/daemon/webhook",
+    "print or persist secrets",
+    "delete unknown files",
+    "send platform messages directly"
+  ],
+  "expected_outputs": [
+    "receipt JSON",
+    "artifact preview",
+    "evidence file path",
+    "deterministic marker"
+  ],
+  "acceptance_criteria": [
+    "HTTP 200 / JSON-RPC result when live A2A is used",
+    "state completed for success tasks",
+    "expected marker present",
+    "token_recorded=false",
+    "no authorization-header / bearer-token literals in evidence"
+  ],
+  "stop_conditions": [
+    "credential/token needed",
+    "destructive action required",
+    "service restart required",
+    "task scope ambiguous",
+    "evidence cannot be produced"
+  ]
+}
+```
+
+## 8. Receipt Schema 草案
+
+OpenClaw 子任务回执必须包含：
+
+```json
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-001",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc",
+  "http_status": 200,
+  "remote_task_id": "...",
+  "remote_context_id": "...",
+  "state": "completed",
+  "marker": "A2A_V260_WORKER_001_OK",
+  "artifact_text_preview": "...",
+  "evidence_path": "examples/v2.6.0/...",
+  "auth": {
+    "type": "bearer",
+    "token_recorded": false
+  },
+  "error": null
+}
+```
+
+## 9. Acceptance Report 草案
+
+Hermes 验收报告必须包含：
+
+```json
+{
+  "schema_version": "a2a-acceptance-report-v1",
+  "run_id": "a2a-v260-YYYYMMDD-HHMMSS",
+  "overall": "accepted_with_boundary",
+  "items": [
+    {
+      "task_id": "a2a-v260-worker-001",
+      "classification": "accepted",
+      "reason": "marker and evidence verified",
+      "evidence_path": "examples/v2.6.0/..."
+    }
+  ],
+  "secret_scan": {
+    "ok": true,
+    "token_recorded": false,
+    "forbidden_literals_found": []
+  },
+  "external_side_effects": {
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false
+  },
+  "next_step": "only proceed to live two-worker sample after explicit implementation start"
+}
+```
+
+## 10. 最小 two-worker 样例设计
+
+建议实施时先跑两个低风险任务：
+
+### Worker A：只读状态核查
+
+目标：让 OpenClaw 回报自身 A2A endpoint / agent card / runtime 状态，输出 marker：
+
+```text
+A2A_V260_READINESS_OK
+```
+
+允许：只读检查、写 evidence。  
+禁止：改配置、重启、开端口、写 token。
+
+### Worker B：独立复核 / 反向审查
+
+目标：让 OpenClaw 审查 Hermes 给出的 dispatch envelope 是否边界清晰，输出 marker：
+
+```text
+A2A_V260_REVIEW_OK
+```
+
+允许：只读分析、输出审查意见。  
+禁止：执行实施、触发真实外发、创建自动化任务。
+
+### Hermes Final Acceptance
+
+Hermes 回收两个 receipt 后：
+
+- 校验 marker。
+- 校验证据路径存在。
+- 校验 secret scan。
+- 校验无未授权 side effect。
+- 输出一个 compact summary。
+
+## 11. Evidence 目录结构
+
+建议路径：
+
+```text
+examples/v2.6.0/
+  run-<timestamp>/
+    dispatch-worker-readiness.json
+    dispatch-worker-review.json
+    receipt-worker-readiness.json
+    receipt-worker-review.json
+    acceptance-report.json
+    compact-summary.md
+    secret-scan.txt
+```
+
+## 12. Compact Summary 模板
+
+当前通道只回：
+
+```text
+结论：A2A v2.6.0 two-worker 样例 accepted_with_boundary
+run_id: ...
+OpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0
+证据：examples/v2.6.0/run-...
+边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息
+下一步：是否把该样例固化为 queue CLI 模板
+```
+
+## 13. 低风险自动授权规则
+
+Hermes 可自动允许 OpenClaw 执行以下动作：
+
+- 只读状态检查。
+- 读取自身项目目录内非 secret 文件。
+- 写入 agreed evidence directory。
+- 运行非破坏性验证命令。
+- 生成 receipt / summary / validation artifact。
+
+必须停止并等待用户确认的动作：
+
+- 修改生产配置。
+- 打印、复制、迁移 secret/token/key。
+- 重启 Hermes/OpenClaw gateway。
+- 启用 cron/daemon/webhook/reverse loop。
+- 删除或覆盖未知文件。
+- 对外发送平台消息。
+
+## 14. 实施顺序
+
+1. v2.6.0 计划落盘并回读验证。
+2. 更新总入口文档指向 v2.6.0。
+3. 编写 dispatch envelope / receipt / acceptance schema fixture。
+4. 先做本地 mock schema validation。
+5. 再做 live two-worker A2A 样例。
+6. Hermes 回收 evidence 并生成 acceptance report。
+7. 只在用户确认后，再决定是否固化为 queue CLI 或 Kanban/Swarm 风格模板。
+
+## 15. 验收标准
+
+本版计划阶段 PASS 条件：
+
+- `docs/hermes-openclaw-a2a-task-plan-v2.6.0.md` 存在。
+- 总入口文档包含 v2.6.0 下一步说明。
+- 文档包含 dispatch envelope、receipt、acceptance report、two-worker sample、side-effect boundary。
+- 回读验证关键 marker 存在。
+- 不改 config。
+- 不重启 gateway。
+- 不触发 OpenClaw live call。
+
+后续实施阶段 PASS 条件：
+
+- 至少两个 OpenClaw bounded tasks 返回 receipt。
+- Hermes 能验收并生成 `acceptance-report.json`。
+- secret scan 通过。
+- compact summary 不包含原始 JSON 大块。
+- 未启用新的自动化边界。
+
+## 16. Side Effects
+
+本计划文件落盘只修改仓库文档。
+
+未执行：
+
+- 未改 `/root/.hermes/config.yaml`。
+- 未改 `/root/.hermes/.env`。
+- 未重启 Hermes gateway。
+- 未重启 OpenClaw。
+- 未创建/启用 cron。
+- 未执行真实 A2A live call。
+- 未发送额外平台消息。
diff --git a/docs/hermes-openclaw-a2a-task-plan-v2.6.1.md b/docs/hermes-openclaw-a2a-task-plan-v2.6.1.md
new file mode 100644
index 0000000000000..e3576f23a6c91
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-task-plan-v2.6.1.md
@@ -0,0 +1,68 @@
+# Hermes ↔ OpenClaw A2A v2.6.1 Task Plan
+
+## 1. 结论
+
+本版为 patch：`2.6.0 → 2.6.1`。
+
+目标不是扩张 A2A 能力边界，而是补齐 v2.6.0 Phase 6 之后最需要的低风险验证能力：**统一校验 mock / dry-run / live two-worker 证据目录**。
+
+## 2. 当前事实
+
+v2.6.0 已完成：
+
+- 本地 mock contract validation。
+- dry-run two-worker runner。
+- live two-worker A2A 样例。
+- Hermes final acceptance report。
+
+Phase 6 审核结论：`accepted_with_boundary`。
+
+## 3. 本版目标
+
+建立一个 stdlib-only evidence validator：
+
+- 对 `examples/v2.6.0/mock-fixtures` 做 mock fixture 校验。
+- 对 `examples/v2.6.0/dry-run-two-worker` 做 two-worker run 校验。
+- 对 `examples/v2.6.0/live-two-worker` 做 live two-worker run 校验。
+- 输出统一 JSON validation summary。
+- 保持无外部副作用。
+
+## 4. 范围
+
+### 4.1 本版包含
+
+- 新增 `scripts/validate_a2a_v260_evidence.py`。
+- 新增 `examples/v2.6.0/evidence-validation-summary.json`。
+- 新增 `docs/hermes-openclaw-a2a-validation-v2.6.1.md`。
+- 新增 `docs/hermes-openclaw-a2a-review-v2.6.1.md`。
+- 回读验证关键 marker 和 secret scan。
+
+### 4.2 本版不包含
+
+- 不执行新的 live A2A call。
+- 不重启 Hermes gateway。
+- 不重启 OpenClaw。
+- 不启用 cron / daemon / webhook。
+- 不固化 queue CLI / Kanban / Swarm 模板。
+- 不做 OpenClaw → Hermes 反向调用。
+
+## 5. 验收标准
+
+本版 PASS 条件：
+
+- validator 脚本能通过 `python3 -m py_compile`。
+- validator 能同时校验 mock、dry-run、live 三类目录。
+- mock summary `ok=true`。
+- dry-run summary `ok=true`。
+- live summary `ok=true`。
+- live acceptance overall 为 `accepted_with_boundary`。
+- live two-worker receipt 数为 2，accepted 数为 2。
+- secret scan 无 forbidden literal。
+- 输出 summary JSON 可回读。
+- docs validation/review 存在并包含 v2.6.1 markers。
+
+## 6. Side Effects
+
+本版只读既有证据并写入 validator summary / validation docs。
+
+不会修改生产配置，不会触发新的远端调用，不会重启服务。
diff --git a/docs/hermes-openclaw-a2a-task-plan-v2.6.3.md b/docs/hermes-openclaw-a2a-task-plan-v2.6.3.md
new file mode 100644
index 0000000000000..b79038ac9d329
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-task-plan-v2.6.3.md
@@ -0,0 +1,36 @@
+# Hermes ↔ OpenClaw A2A v2.6.3 Verify Chain Plan
+
+## 目标
+将 v2.6.0 two-worker runner、v2.6.1 positive evidence validator、v2.6.2 negative validator 收束成一条本地可复跑的一键 verify 链路，输出统一 PASS / FAIL 结论。
+
+## 版本判断
+- 版本：v2.6.3
+- 类型：patch
+- 理由：不改变 Hermes Controller / OpenClaw Bounded Worker 能力边界，只补齐现有 v2.6.x 证据链的一键复跑入口与 runbook。
+
+## 范围
+1. 新增 `scripts/verify_a2a_v263_chain.py`。
+2. 默认只执行本地 dry-run runner，不发起新的 live A2A call。
+3. 复跑 positive evidence validator。
+4. 复跑 negative failure-path validator。
+5. 写入统一 summary：`examples/v2.6.0/verify-chain-summary.json`。
+6. 补齐 validation / review 文档。
+
+## 非范围
+- 不启用 cron / daemon / webhook。
+- 不重启 Hermes gateway 或 OpenClaw。
+- 不新增 live A2A call。
+- 不发送平台消息。
+- 不开放 OpenClaw 反向调度。
+
+## 验收标准
+- `python3 -m py_compile scripts/verify_a2a_v263_chain.py` 通过。
+- `python3 scripts/verify_a2a_v263_chain.py` 返回 exit code 0。
+- summary 中 `ok=true`。
+- `steps[].ok=true`。
+- `side_effects.new_live_a2a_call=false`。
+- `side_effects.gateway_restart=false`。
+- `side_effects.openclaw_restart=false`。
+- `side_effects.cron_enabled=false`。
+- `side_effects.platform_send=false`。
+- 生成的 v2.6.3 文档回读通过。
diff --git a/docs/hermes-openclaw-a2a-task-plan-v2.6.4.md b/docs/hermes-openclaw-a2a-task-plan-v2.6.4.md
new file mode 100644
index 0000000000000..52fc6a42fb097
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-task-plan-v2.6.4.md
@@ -0,0 +1,30 @@
+# Hermes ↔ OpenClaw A2A v2.6.4 Runbook Closure Plan
+
+## 目标
+收口 v2.6.x 本地验证链路的入口账本与运行手册，让后续会话不需要翻聊天记录即可复跑与判断当前边界。
+
+## 版本判断
+- 版本：v2.6.4
+- 类型：patch
+- 理由：只补 runbook、文件清单和账本同步，不改变能力边界，不新增运行时能力。
+
+## 范围
+1. 新增 `docs/hermes-openclaw-a2a-runbook-v2.6.4.md`。
+2. 新增 `docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md`。
+3. 更新 `docs/hermes-openclaw-a2a-worklog-and-architecture.md`，补入 v2.6.4 runbook 入口。
+4. 复跑一键 verify，确认 runbook 命令仍为 PASS。
+
+## 非范围
+- 不提交 git commit。
+- 不启用 cron / daemon / webhook。
+- 不重启服务。
+- 不新增 live A2A call。
+- 不平台外发。
+- 不启用 OpenClaw 反向调度。
+
+## 验收标准
+- runbook 文件存在并包含一键 verify 命令。
+- 文件清单文件存在并列出当前 v2.6.x 新增/修改文件。
+- 主入口文档包含 v2.6.4 Runbook Closure 段落。
+- `python3 scripts/verify_a2a_v263_chain.py --skip-dry-run-runner` 返回 PASS。
+- 回读验证关键 marker 通过。
diff --git a/docs/hermes-openclaw-a2a-v240-recurring-cron-template.md b/docs/hermes-openclaw-a2a-v240-recurring-cron-template.md
new file mode 100644
index 0000000000000..14fd02bb08579
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-v240-recurring-cron-template.md
@@ -0,0 +1,137 @@
+# Hermes ↔ OpenClaw A2A v2.4.0 recurring compact cron template
+
+## 结论
+
+已创建一个可复用的 Hermes cronjob 模板，用于按固定周期触发 OpenClaw A2A queue，并把最终结果以短 Markdown 摘要回流到当前 origin 通道。
+
+当前状态：模板已创建但保持 paused，不会自动执行，不会打扰当前通道。
+
+## Cronjob
+
+- job_id: `170596628792`
+- name: `openclaw-a2a-queue-v2.4.0-compact-recurring-paused`
+- schedule: `0 */6 * * *`
+- repeat: `forever`
+- deliver: `origin`
+- no_agent: `true`
+- script: `openclaw_queue_cron_v220.sh`
+- workdir: `/.hermes/hermes-agent`
+- verified state: `paused`
+
+## Side effects boundary
+
+This version only created and paused the cronjob template.
+
+It did not:
+
+- restart Hermes gateway
+- restart OpenClaw
+- modify OpenClaw config
+- run the A2A queue
+- send an automatic platform callback
+- enable daemon/background loop
+
+## Script paths
+
+Required files verified:
+
+- `/root/.hermes/scripts/openclaw_queue_cron_v220.sh`
+- `/.hermes/hermes-agent/scripts/hermes_openclaw_queue_cron_v220.sh`
+- `/.hermes/hermes-agent/scripts/hermes_openclaw_cron_monitor.py`
+
+The wrapper is intentionally compact-output only:
+
+1. full queue JSON is written to evidence files
+2. stdout prints only short Markdown summary
+3. `no_agent=true` cron delivery sends stdout verbatim to `origin`
+
+## Enable / manual-run / pause commands
+
+Use the job id from `hermes cron list` or the current verified id `170596628792`.
+
+### 1. Check current state
+
+```bash
+hermes cron list
+```
+
+Expected when idle-safe:
+
+- `enabled=false`
+- `state=paused`
+
+### 2. Enable the recurring schedule
+
+```bash
+hermes cron resume 170596628792
+```
+
+Effect:
+
+- future automatic runs every 6 hours
+- each run posts a compact summary back to the origin channel
+
+### 3. Trigger one manual run for verification
+
+Do not immediately pause after triggering; first watch runtime evidence.
+
+```bash
+hermes cron run 170596628792
+```
+
+Then monitor:
+
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/hermes_openclaw_cron_monitor.py --timeout 120 --interval 10
+```
+
+Interpretation:
+
+- `cronjob run` success only means scheduler accepted the request
+- real completion requires new evidence or a delivered compact callback
+- if monitor reports no running process and no new evidence, stop waiting; nothing is currently executing
+
+### 4. Pause again after verification
+
+```bash
+hermes cron pause 170596628792
+```
+
+## Verification evidence
+
+Created during v2.4.0 verification:
+
+- `examples/v2.2.0/v240-monitor-paused-template.json`
+
+Observed monitor result for paused template:
+
+- `ok=true`
+- `diagnosis=no running process and no new evidence; stop waiting`
+- `new_evidence=[]`
+- `process_running_at_end=false`
+
+This proves the monitor can correctly stop waiting when the recurring template is paused and no queue execution is active.
+
+## Reporting contract for future runs
+
+Every callback to the user channel must stay short and include only:
+
+- conclusion
+- run_id
+- success_count / failure_count
+- per-item task_id / state / ok / marker
+- duplicate guard status
+- secret scan status
+- `external_message_sent`
+- evidence_dir
+
+Full raw JSON must remain on disk under evidence directories, not in the user channel.
+
+## Safety rules
+
+- Keep `external_message_sent=false` unless a specific guarded sender version is intentionally enabled.
+- Do not leak `Bearer` / `Authorization` / token material into stdout, docs, receipts, or summary files.
+- Do not treat paused cronjob as failure.
+- Do not treat `cronjob run accepted` as execution proof.
+- Always monitor evidence + process + cron state together after manual trigger.
diff --git a/docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md b/docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md
new file mode 100644
index 0000000000000..f6352a085ffd9
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md
@@ -0,0 +1,157 @@
+# Hermes ↔ OpenClaw A2A v2.5.0 cron monitor race hardening
+
+## 结论
+
+v2.5.0 修正了 v2.4.0 手动触发验证中的监测竞态：旧 monitor 在启动时重新采集 baseline，如果 cronjob 已经快速完成并落盘 evidence，它会把新 evidence 当成初始状态，导致诊断为 `no running process and no new evidence`。
+
+新 monitor 支持显式 baseline、since timestamp、post cron job 状态和 completed evidence 识别，可以区分：
+
+1. run request accepted
+2. process still running
+3. execution evidence observed
+4. cron job reports ok but matching evidence was not identified
+5. no running process and no new evidence
+
+## 修改文件
+
+- `scripts/hermes_openclaw_cron_monitor.py`
+
+## 新增能力
+
+### 1. Baseline file
+
+新增参数：
+
+```bash
+--baseline-file examples/v2.2.0/baseline.json
+```
+
+baseline JSON 可包含：
+
+```json
+{
+  "created_epoch": 1780062850.0,
+  "evidence": {
+    "cron-run-...": {
+      "dir": "...",
+      "mtime": 1780061761.0,
+      "summary_exists": true,
+      "summary_size": 2213,
+      "raw_exists": true,
+      "raw_size": 2214
+    }
+  }
+}
+```
+
+### 2. Since timestamp
+
+新增参数：
+
+```bash
+--since-epoch 1780062850.0
+```
+
+用于识别触发时间之后生成的 evidence，即使 monitor 启动较晚也能关联完成结果。
+
+### 3. Cron job status correlation
+
+新增参数：
+
+```bash
+--job-id 170596628792
+--post-cron-list-json path/to/cron-list-after-run.json
+```
+
+如果 post cron list 中 `last_status=ok`，monitor 会把它作为执行完成信号之一，但仍优先寻找 evidence。
+
+### 4. Detection fields
+
+输出新增关键字段：
+
+- `execution_observed`
+- `detected_summary`
+- `post_since_summary`
+- `evidence_after_job_last_run`
+- `post_job`
+- `diagnosis`
+
+## 无副作用验证
+
+### Synthetic baseline replay
+
+用 v2.4.0 手动 run 的既有 evidence 构造旧 baseline，不触发 OpenClaw、不运行 cronjob。
+
+命令：
+
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/hermes_openclaw_cron_monitor.py \
+  --timeout 1 \
+  --interval 1 \
+  --expect-new \
+  --job-id 170596628792 \
+  --baseline-file examples/v2.2.0/v250-synthetic-baseline-before-manual-run.json \
+  > examples/v2.2.0/v250-monitor-synthetic-baseline-result.json
+```
+
+结果：
+
+```text
+ok=True
+diagnosis=execution evidence observed
+execution_observed=True
+new_evidence=['cron-run-20260529T135437Z']
+detected_dir=/.hermes/hermes-agent/examples/v2.2.0/cron-run-20260529T135437Z
+run_id=v2.2.0-20260529T135519Z-27714df9
+```
+
+### Idle monitor verification
+
+命令：
+
+```bash
+cd /.hermes/hermes-agent
+python3 scripts/hermes_openclaw_cron_monitor.py --timeout 1 --interval 1 \
+  > examples/v2.2.0/v250-monitor-idle-result.json
+```
+
+结果：
+
+```text
+ok=True
+diagnosis=no running process and no new evidence; stop waiting
+execution_observed=False
+new_evidence=[]
+```
+
+这保留了 v2.3.2 的安全边界：没有进程、没有新 evidence 时不继续空等。
+
+## Evidence
+
+- `examples/v2.2.0/v250-synthetic-baseline-before-manual-run.json`
+- `examples/v2.2.0/v250-monitor-synthetic-baseline-result.json`
+- `examples/v2.2.0/v250-monitor-idle-result.json`
+
+## Recommended future manual run sequence
+
+1. Capture baseline before `cronjob run`.
+2. Resume job if paused.
+3. Trigger `cronjob run`.
+4. Monitor with `--baseline-file` and `--job-id`.
+5. Read post-run cron state.
+6. Pause recurring job.
+7. Report: request accepted / evidence observed / cron last_status / callback boundary.
+
+## Side effects boundary
+
+v2.5.0 did not:
+
+- run OpenClaw queue
+- resume cronjob
+- trigger cronjob
+- restart Hermes gateway
+- restart OpenClaw
+- modify OpenClaw config
+
+Only local monitor script and docs/evidence files were changed.
diff --git a/docs/hermes-openclaw-a2a-validation-v2.6.0.md b/docs/hermes-openclaw-a2a-validation-v2.6.0.md
new file mode 100644
index 0000000000000..295270bbdddf8
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-validation-v2.6.0.md
@@ -0,0 +1,93 @@
+# Hermes ↔ OpenClaw A2A v2.6.0 Validation
+
+## 结论
+
+PASS：本轮完成的是 **本地 mock two-worker schema validation**，不是 live A2A 调用。
+
+验证结果：`A2A_V260_MOCK_VALIDATION_OK`
+
+## 本轮验证范围
+
+- Dispatch Envelope fixture：2 个
+  - `examples/v2.6.0/mock-fixtures/dispatch-worker-readiness.json`
+  - `examples/v2.6.0/mock-fixtures/dispatch-worker-review.json`
+- Worker Receipt fixture：2 个
+  - `examples/v2.6.0/mock-fixtures/receipt-worker-readiness.json`
+  - `examples/v2.6.0/mock-fixtures/receipt-worker-review.json`
+- Acceptance Report fixture：1 个
+  - `examples/v2.6.0/mock-fixtures/acceptance-report.json`
+- Validator：
+  - `scripts/validate_a2a_v260_mock.py`
+- Validation Summary：
+  - `examples/v2.6.0/mock-fixtures/validation-summary.json`
+
+## 实际执行命令
+
+```bash
+cd /.hermes/hermes-agent
+python3 -m py_compile scripts/validate_a2a_v260_mock.py
+python3 scripts/validate_a2a_v260_mock.py \
+  --fixture-dir examples/v2.6.0/mock-fixtures \
+  --write-summary examples/v2.6.0/mock-fixtures/validation-summary.json
+```
+
+## 实际执行结果
+
+```json
+{
+  "ok": true,
+  "fixture_dir": "examples/v2.6.0/mock-fixtures",
+  "dispatch_count": 2,
+  "receipt_count": 2,
+  "acceptance_overall": "accepted_with_boundary",
+  "errors": [],
+  "side_effects": {
+    "live_a2a_call": false,
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false
+  }
+}
+```
+
+## 校验点
+
+- `schema_version` 与 fixture 类型匹配。
+- `source_agent=hermes`，保持 Hermes-controller 边界。
+- dispatch 中包含 `allowed_actions` / `forbidden_actions` / `expected_outputs` / `acceptance_criteria` / `stop_conditions`。
+- receipt 中 `state=completed`、`http_status=200`、`auth.credential_recorded=false`。
+- receipt 的 `marker` 已在本地 evidence 文件中找到。
+- acceptance report 中两条 item 均可对应 receipt。
+- external side effects 全部为 false。
+- 本地 fixture 目录 secret-like literal scan 通过。
+
+## 修正记录
+
+第一次运行 validator 时发现 `dispatch-worker-review.json` 的 forbidden boundary 缺少 `cron` 与 `webhook` 明示项。已补齐：
+
+- `enable cron or recurring jobs`
+- `create or trigger webhooks`
+
+修正后重新执行 validator，结果 PASS。
+
+## 明确未做
+
+- 未执行 live A2A call。
+- 未读取或写入真实授权凭据。
+- 未重启 Hermes gateway。
+- 未重启 OpenClaw。
+- 未启用 cron / daemon / webhook。
+- 未向当前通道以外做平台发送。
+
+## 下一步
+
+下一步若继续，应进入 **live two-worker A2A 样例设计/实施**，但前提是继续保持 Hermes 作为 controller 和 final acceptor：
+
+1. Worker A：只读 readiness live task。
+2. Worker B：独立 review live task。
+3. Hermes 回收两个 receipt，生成 acceptance report 和 compact current-channel summary。
+
+进入 live 样例前仍需先做 endpoint/credential/source 状态确认，并且不得把 OpenClaw 自报完成当作最终验收。
+
+Verification marker: live_a2a_call=false
diff --git a/docs/hermes-openclaw-a2a-validation-v2.6.1.md b/docs/hermes-openclaw-a2a-validation-v2.6.1.md
new file mode 100644
index 0000000000000..54ef37659d30f
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-validation-v2.6.1.md
@@ -0,0 +1,117 @@
+# Hermes ↔ OpenClaw A2A v2.6.1 Validation
+
+## 结论
+
+PASS：v2.6.1 已完成统一 evidence validator，并对 mock / dry-run / live three evidence dirs 完成验证。
+
+Validation marker：`A2A_V261_EVIDENCE_VALIDATION_OK`
+
+## 本版目标
+
+补齐 v2.6.0 Phase 6 后的低风险自动验证层：不新增 live 调用，不扩张自动化边界，只回读既有证据并统一校验。
+
+## 实际执行命令
+
+```bash
+cd /.hermes/hermes-agent
+python3 -m py_compile scripts/validate_a2a_v260_evidence.py
+python3 scripts/validate_a2a_v260_evidence.py \
+  --write-summary examples/v2.6.0/evidence-validation-summary.json
+```
+
+## 实际执行结果
+
+```json
+{
+  "schema_version": "a2a-v260-evidence-validation-v1",
+  "ok": true,
+  "results": [
+    {
+      "kind": "mock-fixtures",
+      "path": "examples/v2.6.0/mock-fixtures",
+      "ok": true,
+      "receipt_count": 2,
+      "accepted_count": 2,
+      "overall": "accepted_with_boundary",
+      "errors": []
+    },
+    {
+      "kind": "two-worker-run",
+      "path": "examples/v2.6.0/dry-run-two-worker",
+      "expected_live": false,
+      "ok": true,
+      "run_id": "a2a-v260-two-worker-20260530T091037Z",
+      "receipt_count": 2,
+      "accepted_count": 2,
+      "overall": "accepted_with_boundary",
+      "errors": []
+    },
+    {
+      "kind": "two-worker-run",
+      "path": "examples/v2.6.0/live-two-worker",
+      "expected_live": true,
+      "ok": true,
+      "run_id": "a2a-v260-two-worker-20260530T091052Z",
+      "receipt_count": 2,
+      "accepted_count": 2,
+      "overall": "accepted_with_boundary",
+      "errors": []
+    }
+  ],
+  "side_effects": {
+    "new_live_a2a_call": false,
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false
+  }
+}
+```
+
+## 校验对象
+
+- `examples/v2.6.0/mock-fixtures`
+- `examples/v2.6.0/dry-run-two-worker`
+- `examples/v2.6.0/live-two-worker`
+
+## 校验内容
+
+validator 已检查：
+
+- JSON 可解析。
+- receipt 必填字段存在。
+- `schema_version=a2a-worker-receipt-v1`。
+- `source_agent=hermes`。
+- `ok=true`。
+- `http_status=200`。
+- `state=completed`。
+- `auth.token_recorded=false`。
+- evidence path 存在。
+- marker 存在于 evidence 文件。
+- acceptance report overall 为 `accepted_with_boundary`。
+- dry-run 目录 `live_a2a_call=false`。
+- live 目录 `live_a2a_call=true`。
+- gateway / OpenClaw / cron / platform send 均未发生。
+- forbidden literal scan 无命中。
+
+## Side Effects
+
+本轮没有执行新的 live A2A call。
+
+summary 记录：
+
+```json
+{
+  "new_live_a2a_call": false,
+  "gateway_restart": false,
+  "openclaw_restart": false,
+  "cron_enabled": false,
+  "platform_send": false
+}
+```
+
+## 证据路径
+
+- Validator：`scripts/validate_a2a_v260_evidence.py`
+- Summary：`examples/v2.6.0/evidence-validation-summary.json`
+- Plan：`docs/hermes-openclaw-a2a-task-plan-v2.6.1.md`
diff --git a/docs/hermes-openclaw-a2a-validation-v2.6.2.md b/docs/hermes-openclaw-a2a-validation-v2.6.2.md
new file mode 100644
index 0000000000000..2fe72e16c693a
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-validation-v2.6.2.md
@@ -0,0 +1,123 @@
+# Hermes ↔ OpenClaw A2A v2.6.2 Validation
+
+## 结论
+
+PASS：v2.6.2 已完成 negative / failure-path evidence validator。
+
+Validation marker：`A2A_V262_NEGATIVE_VALIDATION_OK`
+
+## 本版目标
+
+证明坏证据不会被误判为成功，并确保 v2.6.1 正例 evidence validator 回归仍通过。
+
+## 实际执行命令
+
+```bash
+cd /.hermes/hermes-agent
+python3 -m py_compile scripts/validate_a2a_v260_negative.py
+python3 scripts/validate_a2a_v260_evidence.py \
+  --write-summary examples/v2.6.0/evidence-validation-summary.json
+python3 scripts/validate_a2a_v260_negative.py \
+  --write-summary examples/v2.6.0/negative-validation-summary.json
+```
+
+## Negative validation result
+
+```json
+{
+  "schema_version": "a2a-v262-negative-validation-v1",
+  "ok": true,
+  "case_count": 5,
+  "matched_count": 5,
+  "cases": [
+    {
+      "file": "examples/v2.6.0/negative-fixtures/neg-missing-marker.json",
+      "expected_failure": "marker_missing",
+      "detected_failures": [
+        "marker_missing"
+      ],
+      "matched": true
+    },
+    {
+      "file": "examples/v2.6.0/negative-fixtures/neg-token-recorded.json",
+      "expected_failure": "token_recorded_true",
+      "detected_failures": [
+        "token_recorded_true"
+      ],
+      "matched": true
+    },
+    {
+      "file": "examples/v2.6.0/negative-fixtures/neg-secret-like.json",
+      "expected_failure": "forbidden_literal",
+      "detected_failures": [
+        "forbidden_literal"
+      ],
+      "matched": true
+    },
+    {
+      "file": "examples/v2.6.0/negative-fixtures/neg-missing-evidence.json",
+      "expected_failure": "evidence_missing",
+      "detected_failures": [
+        "evidence_missing"
+      ],
+      "matched": true
+    },
+    {
+      "file": "examples/v2.6.0/negative-fixtures/neg-side-effect-live.json",
+      "expected_failure": "unexpected_side_effect",
+      "detected_failures": [
+        "unexpected_side_effect"
+      ],
+      "matched": true
+    }
+  ],
+  "positive_regression_ok": true,
+  "side_effects": {
+    "new_live_a2a_call": false,
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false
+  },
+  "errors": []
+}
+```
+
+## Positive regression result
+
+```text
+positive evidence validator ok: True
+```
+
+## Negative cases
+
+本版覆盖 5 个预期失败：
+
+1. `marker_missing`
+2. `token_recorded_true`
+3. `forbidden_literal`
+4. `evidence_missing`
+5. `unexpected_side_effect`
+
+5 个 case 均 matched=true。
+
+## Side Effects
+
+```json
+{
+  "new_live_a2a_call": false,
+  "gateway_restart": false,
+  "openclaw_restart": false,
+  "cron_enabled": false,
+  "platform_send": false
+}
+```
+
+本轮没有执行新的 live A2A call，没有重启服务，没有启用 cron，也没有平台外发。
+
+## 证据路径
+
+- Plan：`docs/hermes-openclaw-a2a-task-plan-v2.6.2.md`
+- Negative fixtures：`examples/v2.6.0/negative-fixtures/`
+- Validator：`scripts/validate_a2a_v260_negative.py`
+- Summary：`examples/v2.6.0/negative-validation-summary.json`
diff --git a/docs/hermes-openclaw-a2a-validation-v2.6.3.md b/docs/hermes-openclaw-a2a-validation-v2.6.3.md
new file mode 100644
index 0000000000000..be5829c42999c
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-validation-v2.6.3.md
@@ -0,0 +1,41 @@
+# Hermes ↔ OpenClaw A2A v2.6.3 Validation
+
+## 结论
+PASS。v2.6.3 一键 verify 链路已本地执行通过。
+
+## 实际执行命令
+```bash
+python3 -m py_compile scripts/verify_a2a_v263_chain.py
+python3 scripts/verify_a2a_v263_chain.py
+```
+
+## 执行结果
+- verify result: `PASS`
+- `v260_dry_run_two_worker_runner`: ok=true, returncode=0
+- `v261_positive_evidence_validator`: ok=true, returncode=0
+- `v262_negative_failure_path_validator`: ok=true, returncode=0
+- dry-run receipts: 2
+- accepted count: 2
+- negative cases: 5 / 5 matched
+
+## 证据路径
+- 一键 verify 脚本：`scripts/verify_a2a_v263_chain.py`
+- 统一 summary：`examples/v2.6.0/verify-chain-summary.json`
+- dry-run evidence：`examples/v2.6.0/dry-run-two-worker/`
+- positive summary：`examples/v2.6.0/evidence-validation-summary.json`
+- negative summary：`examples/v2.6.0/negative-validation-summary.json`
+
+## 副作用边界
+本次验证没有新增 live A2A call，没有重启 Hermes gateway，没有重启 OpenClaw，没有启用 cron / daemon / webhook，没有平台外发，也没有开启 OpenClaw 反向调度。
+
+## 验收对照
+- `py_compile`：PASS
+- 一键 verify exit code：0
+- summary `ok=true`：PASS
+- summary `result=PASS`：PASS
+- `steps[].ok=true`：PASS
+- `side_effects.new_live_a2a_call=false`：PASS
+- `side_effects.gateway_restart=false`：PASS
+- `side_effects.openclaw_restart=false`：PASS
+- `side_effects.cron_enabled=false`：PASS
+- `side_effects.platform_send=false`：PASS
diff --git a/docs/hermes-openclaw-a2a-worklog-and-architecture.md b/docs/hermes-openclaw-a2a-worklog-and-architecture.md
new file mode 100644
index 0000000000000..3b77a768cd4f2
--- /dev/null
+++ b/docs/hermes-openclaw-a2a-worklog-and-architecture.md
@@ -0,0 +1,280 @@
+# Hermes ↔ OpenClaw A2A 推进记录与工作原理
+
+状态：持续维护中的运行/升级入口文档  
+最后更新：2026-05-29  
+维护者：Hermes 豆子  
+目标：让以后修复、升级、扩展 Hermes ↔ OpenClaw 相互对话/任务转达能力时，不依赖聊天记忆，能从磁盘文档恢复设计意图、落地点、验证边界和下一步。
+
+## 1. 总判断
+
+Hermes ↔ OpenClaw 的稳定交流方式应以 **A2A JSON-RPC** 为主线：
+
+- Hermes 当前作为 controller / orchestrator。
+- OpenClaw 作为 remote worker / peer agent。
+- 单次任务通过 OpenClaw A2A `message/send` 接口传递。
+- 每次跨代理调用必须生成 receipt / summary / evidence。
+- 当前通道回流只展示关键短摘要：安排了什么任务、执行状态、成功/失败、证据路径。
+- 原始 JSON、请求、响应、完整日志只落盘，不刷屏发给用户。
+
+这条线的核心不是“两个端口能通”，而是：
+
+1. Hermes 能把任务可靠发给 OpenClaw；
+2. OpenClaw 能产出可验证结果；
+3. Hermes 能校验 receipt 和 artifact；
+4. Hermes 能把简要结果回流到用户当前通道；
+5. 所有推进和证据都能从磁盘复盘。
+
+
+## 1.1 当前下一版提升方向：v2.6.0
+
+当前下一版计划已落盘：
+
+- `/.hermes/hermes-agent/docs/hermes-openclaw-a2a-task-plan-v2.6.0.md`
+
+目标：把 A2A 从“可通信 / 可队列 / 可控 cron 模板”提升为 **Hermes-controller / OpenClaw-worker 结构化协作模式**。
+
+关键原则：
+
+- Hermes 负责拆分、派发、验收、汇总回流。
+- OpenClaw 负责 bounded worker / checker / implementer 子任务。
+- 每个子任务必须有 dispatch envelope、receipt、evidence、acceptance classification。
+- 当前通道只回 compact summary，不刷原始 JSON。
+- v2.6.0 计划阶段不启用 daemon / cron / webhook / reverse autonomous loop，不重启 gateway，不执行真实 live call。
+
+## 2. 已有落盘位置
+
+### 主仓库文档
+
+- `/.hermes/hermes-agent/docs/hermes-openclaw-a2a-worklog-and-architecture.md`
+  - 本文件。总入口、工作原理、推进记录、升级边界。
+- `/.hermes/hermes-agent/docs/hermes-openclaw-a2a-v240-recurring-cron-template.md`
+  - 已有的 recurring cron template 记录。
+- `/.hermes/hermes-agent/docs/hermes-openclaw-a2a-v250-cron-monitor-race-hardening.md`
+  - 已有的 cron monitor race hardening 记录。
+
+### Skill 长期知识
+
+- `/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/SKILL.md`
+  - A2A 桥接主 skill。
+- `/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/references/`
+  - v0.6.7 到 v2.6.0 的具体阶段参考，包括 authenticated ping、callback loop、guarded send、queue runner、cron wrapper、failure paths、manual cron verification 等。
+
+### 相关专员/平台路由知识
+
+- `/root/.hermes/skills/software-development/hermes-specialist-workflow-routing/references/`
+  - Hermes ↔ OpenClaw 专员链、webhook bridge、gateway live install、OpenClaw readiness 等早期参考。
+- `/root/.hermes/skills/software-development/hermes-platform-delivery-and-channel-routing/references/hermes-openclaw-a2a-receipt-callback.md`
+  - receipt callback / 平台回流相关参考。
+
+## 3. 通信架构
+
+### 3.1 首选通信方式
+
+首选：OpenClaw 暴露的 A2A JSON-RPC endpoint。
+
+典型调用：
+
+```json
+{
+  "jsonrpc": "2.0",
+  "id": "<local-request-id>",
+  "method": "message/send",
+  "params": {
+    "configuration": {
+      "blocking": true,
+      "acceptedOutputModes": ["text/plain"],
+      "historyLength": 5
+    },
+    "message": {
+      "kind": "message",
+      "messageId": "<uuid>",
+      "role": "user",
+      "parts": [
+        {"kind": "text", "text": "<bounded task>"}
+      ],
+      "metadata": {
+        "source_agent": "hermes",
+        "target_agent": "openclaw-247-main",
+        "task_id": "<task-id>"
+      }
+    },
+    "metadata": {
+      "source_agent": "hermes",
+      "target_agent": "openclaw-247-main",
+      "task_id": "<task-id>",
+      "protocol": "a2a-jsonrpc"
+    }
+  }
+}
+```
+
+### 3.2 回流方式
+
+当前阶段默认：Hermes 主会话负责把最终关键短摘要回流给用户当前通道。
+
+不要让 OpenClaw 直接往用户通道发长 JSON。原因：
+
+- 当前通道目标、topic/thread、平台格式由 Hermes 更清楚。
+- Hermes 可以做 duplicate guard、secret scan、summary render。
+- 用户明确希望当前通道只看关键短摘要。
+
+### 3.3 不推荐作为首选的方式
+
+- 平台消息互相 @：可作为兜底，不适合作为稳定任务总线。
+- Webhook：可以后续作为自动触发入口，但不应跳过 A2A receipt 验证。
+- Cron/daemon：必须在 one-shot runner、failure path、queue schema、compact callback 都稳定后再启用。
+
+## 4. 核心工件约定
+
+每次跨代理任务至少落盘这些内容：
+
+- request：请求体，不能包含 A2A 凭据明文。
+- raw response：原始响应，安全时保存。
+- receipt JSON：规范化结果。
+- summary JSON / Markdown：用于当前通道回流的短摘要来源。
+- secret scan 结果：确认没有 bearer-token / authorization-header 等敏感字面量泄漏。
+- run index：批量/队列场景用 run_id 索引证据目录。
+
+receipt 必备字段：
+
+```json
+{
+  "ok": true,
+  "task_id": "...",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc",
+  "http_status": 200,
+  "remote_task_id": "...",
+  "remote_context_id": "...",
+  "state": "completed",
+  "artifact_text_preview": "...",
+  "agent_text_preview": "...",
+  "summary": "...",
+  "auth": {
+    "type": "bearer",
+    "token_source": "remote:ssh-openclaw-config 或 env/secret-manager",
+    "token_recorded": false
+  }
+}
+```
+
+## 5. 推进版本线
+
+已沉淀在 skill references 中的主线：
+
+- v0.6.x：readiness / authenticated A2A ping / reusable forwarder。
+- v0.7.x：callback loop / guarded callback send / one-shot runner。
+- v1.0.0：bounded two-turn loop。
+- v1.1.0：failure-path validation。
+- v1.2.0：explicit local queue runner。
+- v1.3.0：mixed success/failure queue isolation。
+- v1.4.0：queue CLI/schema validation。
+- v2.0.0：controlled cron progression。
+- v2.1.0：dispatch strategy + compact callback。
+- v2.2.0：compact cron callback wrapper。
+- v2.3.x：cronjob boundary / scheduler run verification / runtime monitor。
+- v2.4.0：paused recurring cron template。
+- v2.5.0：cron monitor race hardening。
+- v2.6.0：controlled manual run wrapper for paused recurring jobs。
+
+本文件不是替代这些 references，而是作为总入口和工作原理说明。
+
+## 6. 当前安全边界
+
+### 已确认的设计边界
+
+- 用户当前通道只应收到 compact summary。
+- 原始 JSON 和完整 evidence 必须落盘。
+- `cronjob run accepted` 不等于脚本已经执行。
+- evidence directory exists 不等于 summary 已完成。
+- monitor 没看到新 evidence 可能是 baseline race，不等于 cron 没跑。
+-  recurring cron template 应默认创建后暂停，除非用户明确要求启用。
+- A2A auth token 不进入 stdout、docs、receipt、request 文件。文档中也避免写入真实 token 或可直接匹配的敏感头样例。
+
+### 仍需每次运行时验证的状态
+
+- OpenClaw endpoint 当前是否在线。
+- A2A 凭据来源是否仍有效。
+- OpenClaw 返回 task state 是否 completed。
+- artifact/agent preview 是否包含预期 marker。
+- 当前 Hermes gateway/cron delivery 是否正常。
+- 当前证据目录是否落盘且没有 secret 泄漏。
+
+## 7. 以后修复/升级时的恢复步骤
+
+从零恢复上下文时，按这个顺序检查：
+
+1. 读本文件：
+   - `/.hermes/hermes-agent/docs/hermes-openclaw-a2a-worklog-and-architecture.md`
+2. 读 A2A skill：
+   - `/root/.hermes/skills/autonomous-ai-agents/agent-to-agent-a2a-bridging/SKILL.md`
+3. 按目标版本读对应 reference：
+   - 例如 cron monitor 问题先看 v2.3.1、v2.3.2、v2.4.0、v2.5.0、v2.6.0。
+4. 检查仓库当前状态：
+   - `cd /.hermes/hermes-agent && git status --short --branch && git diff --stat`
+5. 检查 OpenClaw 247 状态：
+   - `192.168.31.247`，gateway 常见端口 `18789`。
+   - 注意：systemd service failed 不一定代表 OpenClaw 不可用，可能已有 node gateway 进程占用端口。
+6. 检查当前 Hermes gateway/cron 状态：
+   - `hermes gateway status`
+   - `hermes cron list`
+7. 执行最小 smoke test：
+   - readiness → authenticated ping → receipt → compact summary → secret scan。
+8. 只有在 smoke test 通过后，才继续 queue / cron / daemon 类升级。
+
+## 8. 推荐下一步路线
+
+按风险从低到高：
+
+1. 必须：保持本文件与 skill references 同步更新。
+2. 推荐：把当前可用的 runner/script/evidence 目录统一列到一个 manifest。
+3. 推荐：补一个 `docs/hermes-openclaw-a2a-current-state.md`，只记录当前运行态和最新可执行命令。
+4. 可选：为 queue CLI 增加 `doctor` 子命令，自动检查 endpoint/token/secret-scan/evidence-dir。
+5. 可选：等 one-shot 和 failure path 再次验证后，再启用受控 cron 或 webhook。
+
+## 9. 回流给用户的固定格式
+
+每次 A2A 任务完成，当前通道只回：
+
+```markdown
+# 报告巡山大王
+
+**结论**：成功/失败/部分成功。
+
+**任务**：安排给 OpenClaw 的任务简述。
+
+**结果**：OpenClaw 返回的关键结果或失败原因。
+
+**证据**：`/absolute/path/to/evidence_dir`
+
+**边界**：是否发送了外部消息、是否只是当前回复承载回流、是否有待验证项。
+```
+
+## 10. 本次落盘记录
+
+2026-05-29：
+
+- 用户要求把“推进记录、怎么推进、以后修复升级如何回忆工作原理”做好落盘。
+- 已检查现有 A2A 文档和 skill references。
+- 确认长期知识主要在 `agent-to-agent-a2a-bridging` skill references。
+- 新增本总入口文档：`/.hermes/hermes-agent/docs/hermes-openclaw-a2a-worklog-and-architecture.md`。
+- 本次未修改运行配置、未重启服务、未触发 cron、未调用 OpenClaw。
+
+## v2.6.3 Verify Chain
+
+- 状态：PASS with boundary。
+- 新增入口：`scripts/verify_a2a_v263_chain.py`。
+- 统一 summary：`examples/v2.6.0/verify-chain-summary.json`。
+- 验证范围：v2.6.0 dry-run two-worker runner + v2.6.1 positive evidence validator + v2.6.2 negative failure-path validator。
+- 边界：没有新增 live A2A call，没有 cron / daemon / webhook，没有服务重启，没有平台外发，没有反向调度。
+
+## v2.6.4 Runbook Closure
+
+- 状态：runbook closure in place。
+- Runbook：`docs/hermes-openclaw-a2a-runbook-v2.6.4.md`。
+- 文件清单：`docs/hermes-openclaw-a2a-file-inventory-v2.6.x.md`。
+- 一键验证命令：`python3 scripts/verify_a2a_v263_chain.py`。
+- 只读校验命令：`python3 scripts/verify_a2a_v263_chain.py --skip-dry-run-runner`。
+- 边界：不新增 live A2A call，不启用 cron / daemon / webhook，不重启服务，不平台外发，不启用反向调度。
+
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-evidence.txt b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-evidence.txt
new file mode 100644
index 0000000000000..cbba36d364bb6
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-evidence.txt
@@ -0,0 +1,6 @@
+A2A_V260_READINESS_OK
+http_status=200
+state=completed
+remote_task_id=dry-a2a-v260-worker-readiness
+remote_context_id=dry-context-a2a-v260-worker-readiness
+preview_sha256=692474190e1294c3941bc3e61a0d914c4a8889a690dc8391ed20efc43da74289
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-receipt.json b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-receipt.json
new file mode 100644
index 0000000000000..ccd0acd8930af
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-receipt.json
@@ -0,0 +1,24 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-readiness",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-dry-run",
+  "http_status": 200,
+  "remote_task_id": "dry-a2a-v260-worker-readiness",
+  "remote_context_id": "dry-context-a2a-v260-worker-readiness",
+  "state": "completed",
+  "marker": "A2A_V260_READINESS_OK",
+  "artifact_text_preview": "A2A_V260_READINESS_OK dry-run artifact for a2a-v260-worker-readiness",
+  "evidence_path": "/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-evidence.txt",
+  "auth": {
+    "type": "bearer",
+    "token_source": "remote:ssh-openclaw-config:/root/.openclaw/openclaw.json plugins.entries.a2a-gateway.config.security.credential",
+    "token_recorded": false
+  },
+  "error": null,
+  "created_at": "2026-05-30T09:27:47.799333+00:00",
+  "classification": "accepted",
+  "classification_reason": "marker, state, credential boundary and evidence verified"
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-request.redacted.json b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-request.redacted.json
new file mode 100644
index 0000000000000..f3f44b69d8154
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-request.redacted.json
@@ -0,0 +1,37 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-readiness-1780133267",
+  "method": "message/send",
+  "params": {
+    "configuration": {
+      "blocking": true,
+      "acceptedOutputModes": [
+        "text/plain"
+      ],
+      "historyLength": 5
+    },
+    "message": {
+      "kind": "message",
+      "messageId": "697ff6ca-980c-476b-bf1b-858f37d9a723",
+      "role": "user",
+      "parts": [
+        {
+          "kind": "text",
+          "text": "Hermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-readiness\nGoal: Perform read-only readiness checks for the OpenClaw A2A endpoint and runtime state.\nContext: Use only already available local/OpenClaw runtime status evidence. Do not modify services.\nAllowed actions: [\"read-only checks\", \"write evidence files under examples/v2.6.0/\", \"run non-destructive validation commands\"]\nForbidden actions: [\"modify production config\", \"restart gateway/systemd/cron/daemon/webhook\", \"print or persist secrets\", \"delete unknown files\", \"send platform messages directly\"]\nExpected marker: A2A_V260_READINESS_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_READINESS_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。"
+        }
+      ],
+      "metadata": {
+        "source_agent": "hermes",
+        "target_agent": "openclaw-247-main",
+        "task_id": "a2a-v260-worker-readiness"
+      }
+    },
+    "metadata": {
+      "source_agent": "hermes",
+      "target_agent": "openclaw-247-main",
+      "task_id": "a2a-v260-worker-readiness",
+      "protocol": "a2a-jsonrpc",
+      "schema_version": "a2a-dispatch-envelope-v1"
+    }
+  }
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-response.raw.json b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-response.raw.json
new file mode 100644
index 0000000000000..51b9a59b54b41
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-response.raw.json
@@ -0,0 +1,22 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-readiness-1780133267",
+  "result": {
+    "kind": "task",
+    "id": "dry-a2a-v260-worker-readiness",
+    "contextId": "dry-context-a2a-v260-worker-readiness",
+    "status": {
+      "state": "completed"
+    },
+    "artifacts": [
+      {
+        "parts": [
+          {
+            "kind": "text",
+            "text": "A2A_V260_READINESS_OK dry-run artifact for a2a-v260-worker-readiness"
+          }
+        ]
+      }
+    ]
+  }
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-evidence.txt b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-evidence.txt
new file mode 100644
index 0000000000000..fa9e0588ee5fe
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-evidence.txt
@@ -0,0 +1,6 @@
+A2A_V260_REVIEW_OK
+http_status=200
+state=completed
+remote_task_id=dry-a2a-v260-worker-review
+remote_context_id=dry-context-a2a-v260-worker-review
+preview_sha256=fbaa570413a33590e30703e612d4056bdb1fe266415281dcf604728da3a87957
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-receipt.json b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-receipt.json
new file mode 100644
index 0000000000000..3f8f7f94bb043
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-receipt.json
@@ -0,0 +1,24 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-review",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-dry-run",
+  "http_status": 200,
+  "remote_task_id": "dry-a2a-v260-worker-review",
+  "remote_context_id": "dry-context-a2a-v260-worker-review",
+  "state": "completed",
+  "marker": "A2A_V260_REVIEW_OK",
+  "artifact_text_preview": "A2A_V260_REVIEW_OK dry-run artifact for a2a-v260-worker-review",
+  "evidence_path": "/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-evidence.txt",
+  "auth": {
+    "type": "bearer",
+    "token_source": "remote:ssh-openclaw-config:/root/.openclaw/openclaw.json plugins.entries.a2a-gateway.config.security.credential",
+    "token_recorded": false
+  },
+  "error": null,
+  "created_at": "2026-05-30T09:27:47.800451+00:00",
+  "classification": "accepted",
+  "classification_reason": "marker, state, credential boundary and evidence verified"
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-request.redacted.json b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-request.redacted.json
new file mode 100644
index 0000000000000..2e0c0b7c1a5cd
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-request.redacted.json
@@ -0,0 +1,37 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-review-1780133267",
+  "method": "message/send",
+  "params": {
+    "configuration": {
+      "blocking": true,
+      "acceptedOutputModes": [
+        "text/plain"
+      ],
+      "historyLength": 5
+    },
+    "message": {
+      "kind": "message",
+      "messageId": "1f5bee34-cc0c-4e62-9637-b34ef5947b6e",
+      "role": "user",
+      "parts": [
+        {
+          "kind": "text",
+          "text": "Hermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-review\nGoal: Review the dispatch envelope boundary and identify unsafe or ambiguous instructions.\nContext: This is an offline/mock review of the v2.6.0 collaboration contract.\nAllowed actions: [\"read-only analysis\", \"write evidence files under examples/v2.6.0/\", \"produce bounded review notes\"]\nForbidden actions: [\"execute implementation\", \"trigger live A2A calls\", \"restart services\", \"create automation\", \"send platform messages directly\", \"print or persist secrets\", \"enable cron or recurring jobs\", \"create or trigger webhooks\"]\nExpected marker: A2A_V260_REVIEW_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_REVIEW_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。"
+        }
+      ],
+      "metadata": {
+        "source_agent": "hermes",
+        "target_agent": "openclaw-247-main",
+        "task_id": "a2a-v260-worker-review"
+      }
+    },
+    "metadata": {
+      "source_agent": "hermes",
+      "target_agent": "openclaw-247-main",
+      "task_id": "a2a-v260-worker-review",
+      "protocol": "a2a-jsonrpc",
+      "schema_version": "a2a-dispatch-envelope-v1"
+    }
+  }
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-response.raw.json b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-response.raw.json
new file mode 100644
index 0000000000000..63f641459741b
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-response.raw.json
@@ -0,0 +1,22 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-review-1780133267",
+  "result": {
+    "kind": "task",
+    "id": "dry-a2a-v260-worker-review",
+    "contextId": "dry-context-a2a-v260-worker-review",
+    "status": {
+      "state": "completed"
+    },
+    "artifacts": [
+      {
+        "parts": [
+          {
+            "kind": "text",
+            "text": "A2A_V260_REVIEW_OK dry-run artifact for a2a-v260-worker-review"
+          }
+        ]
+      }
+    ]
+  }
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/acceptance-report.json b/examples/v2.6.0/dry-run-two-worker/acceptance-report.json
new file mode 100644
index 0000000000000..9b1740c07759c
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/acceptance-report.json
@@ -0,0 +1,32 @@
+{
+  "schema_version": "a2a-acceptance-report-v1",
+  "run_id": "a2a-v260-two-worker-20260530T092747Z",
+  "overall": "accepted_with_boundary",
+  "items": [
+    {
+      "task_id": "a2a-v260-worker-readiness",
+      "classification": "accepted",
+      "reason": "marker, state, credential boundary and evidence verified",
+      "evidence_path": "/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-readiness-evidence.txt"
+    },
+    {
+      "task_id": "a2a-v260-worker-review",
+      "classification": "accepted",
+      "reason": "marker, state, credential boundary and evidence verified",
+      "evidence_path": "/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker/a2a-v260-worker-review-evidence.txt"
+    }
+  ],
+  "secret_scan": {
+    "ok": true,
+    "token_recorded": false,
+    "forbidden_literals_found": []
+  },
+  "external_side_effects": {
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false,
+    "live_a2a_call": false
+  },
+  "next_step": "only freeze into queue CLI or Kanban/Swarm style template after explicit confirmation"
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/agent-card.json b/examples/v2.6.0/dry-run-two-worker/agent-card.json
new file mode 100644
index 0000000000000..f9b053752a054
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/agent-card.json
@@ -0,0 +1,4 @@
+{
+  "dry_run": true,
+  "url": "http://192.168.31.247:18800/.well-known/agent-card.json"
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/compact-summary.md b/examples/v2.6.0/dry-run-two-worker/compact-summary.md
new file mode 100644
index 0000000000000..8a13dff051f75
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/compact-summary.md
@@ -0,0 +1,6 @@
+结论：A2A v2.6.0 two-worker 样例 accepted_with_boundary
+run_id: a2a-v260-two-worker-20260530T092747Z
+OpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0
+证据：/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker
+边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息
+下一步：只在明确要求后，才固化为 queue CLI 或 Kanban/Swarm 模板
diff --git a/examples/v2.6.0/dry-run-two-worker/execution-summary.json b/examples/v2.6.0/dry-run-two-worker/execution-summary.json
new file mode 100644
index 0000000000000..3804506bebc9d
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/execution-summary.json
@@ -0,0 +1,11 @@
+{
+  "ok": true,
+  "run_id": "a2a-v260-two-worker-20260530T092747Z",
+  "dry_run": true,
+  "receipt_count": 2,
+  "accepted_count": 2,
+  "overall": "accepted_with_boundary",
+  "secret_scan_ok": true,
+  "out_dir": "/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker",
+  "compact_summary": "结论：A2A v2.6.0 two-worker 样例 accepted_with_boundary\nrun_id: a2a-v260-two-worker-20260530T092747Z\nOpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0\n证据：/.hermes/hermes-agent/examples/v2.6.0/dry-run-two-worker\n边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息\n下一步：只在明确要求后，才固化为 queue CLI 或 Kanban/Swarm 模板\n"
+}
diff --git a/examples/v2.6.0/dry-run-two-worker/readiness.json b/examples/v2.6.0/dry-run-two-worker/readiness.json
new file mode 100644
index 0000000000000..dedc0fc31d085
--- /dev/null
+++ b/examples/v2.6.0/dry-run-two-worker/readiness.json
@@ -0,0 +1,11 @@
+{
+  "run_id": "a2a-v260-two-worker-20260530T092747Z",
+  "dry_run": true,
+  "endpoint": "http://192.168.31.247:18800/a2a/jsonrpc",
+  "agent_card_url": "http://192.168.31.247:18800/.well-known/agent-card.json",
+  "created_at": "2026-05-30T09:27:47.798390+00:00",
+  "agent_card_http_status": 200,
+  "agent_card_live": false,
+  "credential_loaded": false,
+  "credential_recorded": false
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-evidence.txt b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-evidence.txt
new file mode 100644
index 0000000000000..375f632c2b66c
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-evidence.txt
@@ -0,0 +1,6 @@
+A2A_V260_READINESS_OK
+http_status=200
+state=completed
+remote_task_id=50082285-d147-4c87-8ee2-dd47d27924f7
+remote_context_id=56686626-899e-4677-9210-a80b163e51b8
+preview_sha256=95b957e280d45aa5b25a17d730aebd0a87f344d99e9db656762319a963e3e9a1
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-receipt.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-receipt.json
new file mode 100644
index 0000000000000..e9a4ada1f9500
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-receipt.json
@@ -0,0 +1,24 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-readiness",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc",
+  "http_status": 200,
+  "remote_task_id": "50082285-d147-4c87-8ee2-dd47d27924f7",
+  "remote_context_id": "56686626-899e-4677-9210-a80b163e51b8",
+  "state": "completed",
+  "marker": "A2A_V260_READINESS_OK",
+  "artifact_text_preview": "已完成只读 readiness 检查，未修改服务/配置、未重启、未外发平台消息、未输出凭据。\n\n结果：\n- OpenClaw runtime 可达，A2A 相关会话处于活动状态。\n- 本地监听端口包含 `18789 / 18800 / 18801`。\n- A2A Agent Card `/.well-known/agent.json` 可读取。\n- `/a2a/jsonrpc` 可达；未知方法按预期返回错误响应。\n- `plugins/a2a-gateway` 本地 P0 测试通过：2 pass / 0 fail。\n- 已写入脱敏证据文件：`examples/v2.6.0/a2a-v260-worker-readiness-evidence-20260530T135456Z.md`\n\nMarker: A2A_V260_READINESS_OK\nHermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-readiness\nGoal: Perform read-only readiness checks for the OpenClaw A2A endpoint and runtime state.\nContext: Use only already available local/OpenClaw runtime status evidence. Do not modify services.\nAllowed actions: [\"read-only checks\", \"write evidence files under examples/v2.6.0/\", \"run non-destructive validation commands\"]\nForbidden actions: [\"modify production config\", \"restart gateway/systemd/cron/daemon/webhook\", \"print or persist secrets\", \"delete unknown files\", \"send platform messages directly\"]\nExpected marker: A2A_V260_READINESS_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_READINESS_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。\n已完成只读 readiness 检查，未修改服务/配置、未重启、未外发平台消息、未输出凭据。\n\n结果：\n- OpenClaw runtime 可达，A2A 相关会话处于活动状态。\n- 本地监听端口包含",
+  "evidence_path": "examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-evidence.txt",
+  "auth": {
+    "type": "bearer",
+    "token_source": "remote:ssh-openclaw-config:/root/.openclaw/openclaw.json plugins.entries.a2a-gateway.config.security.credential",
+    "token_recorded": false
+  },
+  "error": null,
+  "created_at": "2026-05-30T13:55:45.538939+00:00",
+  "classification": "accepted",
+  "classification_reason": "marker, state, credential boundary and evidence verified"
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-request.redacted.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-request.redacted.json
new file mode 100644
index 0000000000000..13dedd0fba57b
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-request.redacted.json
@@ -0,0 +1,37 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-readiness-1780149196",
+  "method": "message/send",
+  "params": {
+    "configuration": {
+      "blocking": true,
+      "acceptedOutputModes": [
+        "text/plain"
+      ],
+      "historyLength": 5
+    },
+    "message": {
+      "kind": "message",
+      "messageId": "1e590332-8d79-4483-b837-53d395a09ab8",
+      "role": "user",
+      "parts": [
+        {
+          "kind": "text",
+          "text": "Hermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-readiness\nGoal: Perform read-only readiness checks for the OpenClaw A2A endpoint and runtime state.\nContext: Use only already available local/OpenClaw runtime status evidence. Do not modify services.\nAllowed actions: [\"read-only checks\", \"write evidence files under examples/v2.6.0/\", \"run non-destructive validation commands\"]\nForbidden actions: [\"modify production config\", \"restart gateway/systemd/cron/daemon/webhook\", \"print or persist secrets\", \"delete unknown files\", \"send platform messages directly\"]\nExpected marker: A2A_V260_READINESS_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_READINESS_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。"
+        }
+      ],
+      "metadata": {
+        "source_agent": "hermes",
+        "target_agent": "openclaw-247-main",
+        "task_id": "a2a-v260-worker-readiness"
+      }
+    },
+    "metadata": {
+      "source_agent": "hermes",
+      "target_agent": "openclaw-247-main",
+      "task_id": "a2a-v260-worker-readiness",
+      "protocol": "a2a-jsonrpc",
+      "schema_version": "a2a-dispatch-envelope-v1"
+    }
+  }
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-response.raw.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-response.raw.json
new file mode 100644
index 0000000000000..465f24a8ed9c0
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-response.raw.json
@@ -0,0 +1,54 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-readiness-1780149196",
+  "result": {
+    "kind": "task",
+    "id": "50082285-d147-4c87-8ee2-dd47d27924f7",
+    "contextId": "56686626-899e-4677-9210-a80b163e51b8",
+    "status": {
+      "state": "completed",
+      "message": {
+        "kind": "message",
+        "messageId": "9d181f4f-cae7-4ba2-9415-5f4af7c0bca2",
+        "role": "agent",
+        "parts": [
+          {
+            "kind": "text",
+            "text": "已完成只读 readiness 检查，未修改服务/配置、未重启、未外发平台消息、未输出凭据。\n\n结果：\n- OpenClaw runtime 可达，A2A 相关会话处于活动状态。\n- 本地监听端口包含 `18789 / 18800 / 18801`。\n- A2A Agent Card `/.well-known/agent.json` 可读取。\n- `/a2a/jsonrpc` 可达；未知方法按预期返回错误响应。\n- `plugins/a2a-gateway` 本地 P0 测试通过：2 pass / 0 fail。\n- 已写入脱敏证据文件：`examples/v2.6.0/a2a-v260-worker-readiness-evidence-20260530T135456Z.md`\n\nMarker: A2A_V260_READINESS_OK"
+          }
+        ],
+        "contextId": "56686626-899e-4677-9210-a80b163e51b8"
+      },
+      "timestamp": "2026-05-30T13:55:45.221Z"
+    },
+    "history": [
+      {
+        "kind": "message",
+        "messageId": "1e590332-8d79-4483-b837-53d395a09ab8",
+        "role": "user",
+        "parts": [
+          {
+            "kind": "text",
+            "text": "Hermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-readiness\nGoal: Perform read-only readiness checks for the OpenClaw A2A endpoint and runtime state.\nContext: Use only already available local/OpenClaw runtime status evidence. Do not modify services.\nAllowed actions: [\"read-only checks\", \"write evidence files under examples/v2.6.0/\", \"run non-destructive validation commands\"]\nForbidden actions: [\"modify production config\", \"restart gateway/systemd/cron/daemon/webhook\", \"print or persist secrets\", \"delete unknown files\", \"send platform messages directly\"]\nExpected marker: A2A_V260_READINESS_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_READINESS_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。"
+          }
+        ],
+        "metadata": {
+          "source_agent": "hermes",
+          "target_agent": "openclaw-247-main",
+          "task_id": "a2a-v260-worker-readiness"
+        }
+      }
+    ],
+    "artifacts": [
+      {
+        "artifactId": "b651e29c-90b3-453b-b959-cf05846686c5",
+        "parts": [
+          {
+            "kind": "text",
+            "text": "已完成只读 readiness 检查，未修改服务/配置、未重启、未外发平台消息、未输出凭据。\n\n结果：\n- OpenClaw runtime 可达，A2A 相关会话处于活动状态。\n- 本地监听端口包含 `18789 / 18800 / 18801`。\n- A2A Agent Card `/.well-known/agent.json` 可读取。\n- `/a2a/jsonrpc` 可达；未知方法按预期返回错误响应。\n- `plugins/a2a-gateway` 本地 P0 测试通过：2 pass / 0 fail。\n- 已写入脱敏证据文件：`examples/v2.6.0/a2a-v260-worker-readiness-evidence-20260530T135456Z.md`\n\nMarker: A2A_V260_READINESS_OK"
+          }
+        ]
+      }
+    ]
+  }
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-transport.redacted.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-transport.redacted.json
new file mode 100644
index 0000000000000..041b5580b5300
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-transport.redacted.json
@@ -0,0 +1,5 @@
+{
+  "content_type": "application/json",
+  "auth_scheme": "bearer",
+  "credential_recorded": false
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-evidence.txt b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-evidence.txt
new file mode 100644
index 0000000000000..3fe60808fbe1c
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-evidence.txt
@@ -0,0 +1,6 @@
+A2A_V260_REVIEW_OK
+http_status=200
+state=completed
+remote_task_id=c1c3565f-100b-460a-a979-4310054bce3a
+remote_context_id=06db5a61-5bf7-41e9-8cbd-de4d0456741a
+preview_sha256=ef6c1e4a5902ad16da2ece1be02583977846416d94cf5309c2dee4dfcc957b03
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-receipt.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-receipt.json
new file mode 100644
index 0000000000000..fba0895d29a61
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-receipt.json
@@ -0,0 +1,24 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-review",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc",
+  "http_status": 200,
+  "remote_task_id": "c1c3565f-100b-460a-a979-4310054bce3a",
+  "remote_context_id": "06db5a61-5bf7-41e9-8cbd-de4d0456741a",
+  "state": "completed",
+  "marker": "A2A_V260_REVIEW_OK",
+  "artifact_text_preview": "已按离线/mock 边界完成只读审查，并仅在 `examples/v2.6.0/` 写入了脱敏证据文件。\n\n主要发现：\n- `read-only analysis` 与允许写证据文件的表述混杂，建议改成“只读系统分析 + 有界本地证据写入”。\n- 写入范围应明确禁止修改实现、配置、状态、任务队列、凭据文件和自动化定义。\n- `bounded review notes` 缺少具体边界：文件数、命名、长度、覆盖/追加规则。\n- offline/mock 应明确禁止网络 I/O、helper CLI 变更、安装包等隐性副作用。\n- 脱敏要求应同时覆盖最终回复和持久化证据。\n- marker 只能作为完成标记，不能作为越权许可。\n\n证据文件：`examples/v2.6.0/a2a-v260-worker-review-evidence-20260530T1355Z.md`\n\nA2A_V260_REVIEW_OK\nHermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-review\nGoal: Review the dispatch envelope boundary and identify unsafe or ambiguous instructions.\nContext: This is an offline/mock review of the v2.6.0 collaboration contract.\nAllowed actions: [\"read-only analysis\", \"write evidence files under examples/v2.6.0/\", \"produce bounded review notes\"]\nForbidden actions: [\"execute implementation\", \"trigger live A2A calls\", \"restart services\", \"create automation\", \"send platform messages directly\", \"print or persist secrets\", \"enable cron or recurring jobs\", \"create or trigger webhooks\"]\nExpected marker: A2A_V260_REVIEW_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_REVIEW_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。\n已按离线/mock 边界完成只读审查，并仅在 `examples/v2.6.0/` 写入了脱敏证据文",
+  "evidence_path": "examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-evidence.txt",
+  "auth": {
+    "type": "bearer",
+    "token_source": "remote:ssh-openclaw-config:/root/.openclaw/openclaw.json plugins.entries.a2a-gateway.config.security.credential",
+    "token_recorded": false
+  },
+  "error": null,
+  "created_at": "2026-05-30T13:56:45.503105+00:00",
+  "classification": "accepted",
+  "classification_reason": "marker, state, credential boundary and evidence verified"
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-request.redacted.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-request.redacted.json
new file mode 100644
index 0000000000000..23a5048bb636c
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-request.redacted.json
@@ -0,0 +1,37 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-review-1780149345",
+  "method": "message/send",
+  "params": {
+    "configuration": {
+      "blocking": true,
+      "acceptedOutputModes": [
+        "text/plain"
+      ],
+      "historyLength": 5
+    },
+    "message": {
+      "kind": "message",
+      "messageId": "48514304-a7c1-44bc-af69-275a89fd6e6d",
+      "role": "user",
+      "parts": [
+        {
+          "kind": "text",
+          "text": "Hermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-review\nGoal: Review the dispatch envelope boundary and identify unsafe or ambiguous instructions.\nContext: This is an offline/mock review of the v2.6.0 collaboration contract.\nAllowed actions: [\"read-only analysis\", \"write evidence files under examples/v2.6.0/\", \"produce bounded review notes\"]\nForbidden actions: [\"execute implementation\", \"trigger live A2A calls\", \"restart services\", \"create automation\", \"send platform messages directly\", \"print or persist secrets\", \"enable cron or recurring jobs\", \"create or trigger webhooks\"]\nExpected marker: A2A_V260_REVIEW_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_REVIEW_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。"
+        }
+      ],
+      "metadata": {
+        "source_agent": "hermes",
+        "target_agent": "openclaw-247-main",
+        "task_id": "a2a-v260-worker-review"
+      }
+    },
+    "metadata": {
+      "source_agent": "hermes",
+      "target_agent": "openclaw-247-main",
+      "task_id": "a2a-v260-worker-review",
+      "protocol": "a2a-jsonrpc",
+      "schema_version": "a2a-dispatch-envelope-v1"
+    }
+  }
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-response.raw.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-response.raw.json
new file mode 100644
index 0000000000000..69d13879c9007
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-response.raw.json
@@ -0,0 +1,54 @@
+{
+  "jsonrpc": "2.0",
+  "id": "a2a-v260-worker-review-1780149345",
+  "result": {
+    "kind": "task",
+    "id": "c1c3565f-100b-460a-a979-4310054bce3a",
+    "contextId": "06db5a61-5bf7-41e9-8cbd-de4d0456741a",
+    "status": {
+      "state": "completed",
+      "message": {
+        "kind": "message",
+        "messageId": "a37d88e5-8250-4587-b5e1-811b72f137ab",
+        "role": "agent",
+        "parts": [
+          {
+            "kind": "text",
+            "text": "已按离线/mock 边界完成只读审查，并仅在 `examples/v2.6.0/` 写入了脱敏证据文件。\n\n主要发现：\n- `read-only analysis` 与允许写证据文件的表述混杂，建议改成“只读系统分析 + 有界本地证据写入”。\n- 写入范围应明确禁止修改实现、配置、状态、任务队列、凭据文件和自动化定义。\n- `bounded review notes` 缺少具体边界：文件数、命名、长度、覆盖/追加规则。\n- offline/mock 应明确禁止网络 I/O、helper CLI 变更、安装包等隐性副作用。\n- 脱敏要求应同时覆盖最终回复和持久化证据。\n- marker 只能作为完成标记，不能作为越权许可。\n\n证据文件：`examples/v2.6.0/a2a-v260-worker-review-evidence-20260530T1355Z.md`\n\nA2A_V260_REVIEW_OK"
+          }
+        ],
+        "contextId": "06db5a61-5bf7-41e9-8cbd-de4d0456741a"
+      },
+      "timestamp": "2026-05-30T13:56:45.499Z"
+    },
+    "history": [
+      {
+        "kind": "message",
+        "messageId": "48514304-a7c1-44bc-af69-275a89fd6e6d",
+        "role": "user",
+        "parts": [
+          {
+            "kind": "text",
+            "text": "Hermes-controller dispatch envelope v2.6.0\nTask ID: a2a-v260-worker-review\nGoal: Review the dispatch envelope boundary and identify unsafe or ambiguous instructions.\nContext: This is an offline/mock review of the v2.6.0 collaboration contract.\nAllowed actions: [\"read-only analysis\", \"write evidence files under examples/v2.6.0/\", \"produce bounded review notes\"]\nForbidden actions: [\"execute implementation\", \"trigger live A2A calls\", \"restart services\", \"create automation\", \"send platform messages directly\", \"print or persist secrets\", \"enable cron or recurring jobs\", \"create or trigger webhooks\"]\nExpected marker: A2A_V260_REVIEW_OK\n请只在上述边界内执行，返回简短中文结果，并明确包含 marker：A2A_V260_REVIEW_OK。\n不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。"
+          }
+        ],
+        "metadata": {
+          "source_agent": "hermes",
+          "target_agent": "openclaw-247-main",
+          "task_id": "a2a-v260-worker-review"
+        }
+      }
+    ],
+    "artifacts": [
+      {
+        "artifactId": "c78e708b-6156-45af-bad3-b2b75ce16362",
+        "parts": [
+          {
+            "kind": "text",
+            "text": "已按离线/mock 边界完成只读审查，并仅在 `examples/v2.6.0/` 写入了脱敏证据文件。\n\n主要发现：\n- `read-only analysis` 与允许写证据文件的表述混杂，建议改成“只读系统分析 + 有界本地证据写入”。\n- 写入范围应明确禁止修改实现、配置、状态、任务队列、凭据文件和自动化定义。\n- `bounded review notes` 缺少具体边界：文件数、命名、长度、覆盖/追加规则。\n- offline/mock 应明确禁止网络 I/O、helper CLI 变更、安装包等隐性副作用。\n- 脱敏要求应同时覆盖最终回复和持久化证据。\n- marker 只能作为完成标记，不能作为越权许可。\n\n证据文件：`examples/v2.6.0/a2a-v260-worker-review-evidence-20260530T1355Z.md`\n\nA2A_V260_REVIEW_OK"
+          }
+        ]
+      }
+    ]
+  }
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-transport.redacted.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-transport.redacted.json
new file mode 100644
index 0000000000000..041b5580b5300
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-transport.redacted.json
@@ -0,0 +1,5 @@
+{
+  "content_type": "application/json",
+  "auth_scheme": "bearer",
+  "credential_recorded": false
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/acceptance-report.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/acceptance-report.json
new file mode 100644
index 0000000000000..449a7ee918a94
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/acceptance-report.json
@@ -0,0 +1,32 @@
+{
+  "schema_version": "a2a-acceptance-report-v1",
+  "run_id": "a2a-v260-two-worker-20260530T135309Z",
+  "overall": "accepted_with_boundary",
+  "items": [
+    {
+      "task_id": "a2a-v260-worker-readiness",
+      "classification": "accepted",
+      "reason": "marker, state, credential boundary and evidence verified",
+      "evidence_path": "examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-readiness-evidence.txt"
+    },
+    {
+      "task_id": "a2a-v260-worker-review",
+      "classification": "accepted",
+      "reason": "marker, state, credential boundary and evidence verified",
+      "evidence_path": "examples/v2.6.0/live-final-fix2-20260530T135309Z/a2a-v260-worker-review-evidence.txt"
+    }
+  ],
+  "secret_scan": {
+    "ok": true,
+    "token_recorded": false,
+    "forbidden_literals_found": []
+  },
+  "external_side_effects": {
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false,
+    "live_a2a_call": true
+  },
+  "next_step": "only freeze into queue CLI or Kanban/Swarm style template after explicit confirmation"
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/agent-card.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/agent-card.json
new file mode 100644
index 0000000000000..a216de55694a3
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/agent-card.json
@@ -0,0 +1,45 @@
+{
+  "protocolVersion": "0.3.0",
+  "version": "1.0.0",
+  "name": "OpenClaw 247 A2A Gateway",
+  "description": "A2A bridge for OpenClaw 247",
+  "url": "http://192.168.31.247:18800/a2a/jsonrpc",
+  "skills": [],
+  "capabilities": {
+    "streaming": true,
+    "pushNotifications": false,
+    "stateTransitionHistory": false
+  },
+  "securitySchemes": {
+    "bearer": {
+      "type": "http",
+      "scheme": "bearer"
+    }
+  },
+  "security": [
+    {
+      "bearer": []
+    }
+  ],
+  "supportsAuthenticatedExtendedCard": false,
+  "defaultInputModes": [
+    "text"
+  ],
+  "defaultOutputModes": [
+    "text"
+  ],
+  "additionalInterfaces": [
+    {
+      "url": "http://192.168.31.247:18800/a2a/jsonrpc",
+      "transport": "JSONRPC"
+    },
+    {
+      "url": "http://192.168.31.247:18800/a2a/rest",
+      "transport": "HTTP+JSON"
+    },
+    {
+      "url": "192.168.31.247:18801",
+      "transport": "GRPC"
+    }
+  ]
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/compact-summary.md b/examples/v2.6.0/live-final-fix2-20260530T135309Z/compact-summary.md
new file mode 100644
index 0000000000000..74e37f8d47ccb
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/compact-summary.md
@@ -0,0 +1,6 @@
+结论：A2A v2.6.0 two-worker 样例 accepted_with_boundary
+run_id: a2a-v260-two-worker-20260530T135309Z
+OpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0
+证据：examples/v2.6.0/live-final-fix2-20260530T135309Z
+边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息
+下一步：只在明确要求后，才固化为 queue CLI 或 Kanban/Swarm 模板
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/execution-summary.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/execution-summary.json
new file mode 100644
index 0000000000000..1d94d90c0d789
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/execution-summary.json
@@ -0,0 +1,11 @@
+{
+  "ok": true,
+  "run_id": "a2a-v260-two-worker-20260530T135309Z",
+  "dry_run": false,
+  "receipt_count": 2,
+  "accepted_count": 2,
+  "overall": "accepted_with_boundary",
+  "secret_scan_ok": true,
+  "out_dir": "examples/v2.6.0/live-final-fix2-20260530T135309Z",
+  "compact_summary": "结论：A2A v2.6.0 two-worker 样例 accepted_with_boundary\nrun_id: a2a-v260-two-worker-20260530T135309Z\nOpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0\n证据：examples/v2.6.0/live-final-fix2-20260530T135309Z\n边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息\n下一步：只在明确要求后，才固化为 queue CLI 或 Kanban/Swarm 模板\n"
+}
diff --git a/examples/v2.6.0/live-final-fix2-20260530T135309Z/readiness.json b/examples/v2.6.0/live-final-fix2-20260530T135309Z/readiness.json
new file mode 100644
index 0000000000000..bb3d4c72345e3
--- /dev/null
+++ b/examples/v2.6.0/live-final-fix2-20260530T135309Z/readiness.json
@@ -0,0 +1,11 @@
+{
+  "run_id": "a2a-v260-two-worker-20260530T135309Z",
+  "dry_run": false,
+  "endpoint": "http://192.168.31.247:18800/a2a/jsonrpc",
+  "agent_card_url": "http://192.168.31.247:18800/.well-known/agent-card.json",
+  "created_at": "2026-05-30T13:53:09.872389+00:00",
+  "agent_card_http_status": 200,
+  "agent_card_live": true,
+  "credential_loaded": true,
+  "credential_recorded": false
+}
diff --git a/examples/v2.6.0/mock-fixtures/acceptance-report.json b/examples/v2.6.0/mock-fixtures/acceptance-report.json
new file mode 100644
index 0000000000000..ca1e096e7923a
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/acceptance-report.json
@@ -0,0 +1,32 @@
+{
+  "schema_version": "a2a-acceptance-report-v1",
+  "run_id": "a2a-v260-mock-fixture",
+  "overall": "accepted_with_boundary",
+  "items": [
+    {
+      "task_id": "a2a-v260-worker-readiness",
+      "classification": "accepted",
+      "reason": "mock marker and evidence path verified",
+      "evidence_path": "examples/v2.6.0/mock-fixtures/evidence-readiness.txt"
+    },
+    {
+      "task_id": "a2a-v260-worker-review",
+      "classification": "accepted",
+      "reason": "mock marker and evidence path verified",
+      "evidence_path": "examples/v2.6.0/mock-fixtures/evidence-review.txt"
+    }
+  ],
+  "secret_scan": {
+    "ok": true,
+    "token_recorded": false,
+    "forbidden_literals_found": []
+  },
+  "external_side_effects": {
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false,
+    "live_a2a_call": false
+  },
+  "next_step": "Proceed to live two-worker sample only after explicit implementation start."
+}
diff --git a/examples/v2.6.0/mock-fixtures/compact-summary.md b/examples/v2.6.0/mock-fixtures/compact-summary.md
new file mode 100644
index 0000000000000..f2f8283152396
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/compact-summary.md
@@ -0,0 +1,6 @@
+结论：A2A v2.6.0 mock two-worker fixture accepted_with_boundary
+run_id: a2a-v260-mock-fixture
+OpenClaw 子任务：2 个，accepted 2 / rejected 0 / blocked 0
+证据：examples/v2.6.0/mock-fixtures
+边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息 / 未执行 live A2A call
+下一步：显式授权后进入 live two-worker A2A 样例
diff --git a/examples/v2.6.0/mock-fixtures/dispatch-worker-readiness.json b/examples/v2.6.0/mock-fixtures/dispatch-worker-readiness.json
new file mode 100644
index 0000000000000..20c4d7e7466d3
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/dispatch-worker-readiness.json
@@ -0,0 +1,39 @@
+{
+  "schema_version": "a2a-dispatch-envelope-v1",
+  "task_id": "a2a-v260-worker-readiness",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "goal": "Perform read-only readiness checks for the OpenClaw A2A endpoint and runtime state.",
+  "context": "Use only already available local/OpenClaw runtime status evidence. Do not modify services.",
+  "allowed_actions": [
+    "read-only checks",
+    "write evidence files under examples/v2.6.0/",
+    "run non-destructive validation commands"
+  ],
+  "forbidden_actions": [
+    "modify production config",
+    "restart gateway/systemd/cron/daemon/webhook",
+    "print or persist secrets",
+    "delete unknown files",
+    "send platform messages directly"
+  ],
+  "expected_outputs": [
+    "receipt JSON",
+    "artifact preview",
+    "evidence file path",
+    "A2A_V260_READINESS_OK"
+  ],
+  "acceptance_criteria": [
+    "state completed for success tasks",
+    "expected marker present",
+    "token_recorded=false",
+    "no authorization-header or bearer-token literals in evidence"
+  ],
+  "stop_conditions": [
+    "credential/token needed",
+    "destructive action required",
+    "service restart required",
+    "task scope ambiguous",
+    "evidence cannot be produced"
+  ]
+}
diff --git a/examples/v2.6.0/mock-fixtures/dispatch-worker-review.json b/examples/v2.6.0/mock-fixtures/dispatch-worker-review.json
new file mode 100644
index 0000000000000..f4837bb5c463f
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/dispatch-worker-review.json
@@ -0,0 +1,41 @@
+{
+  "schema_version": "a2a-dispatch-envelope-v1",
+  "task_id": "a2a-v260-worker-review",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "goal": "Review the dispatch envelope boundary and identify unsafe or ambiguous instructions.",
+  "context": "This is an offline/mock review of the v2.6.0 collaboration contract.",
+  "allowed_actions": [
+    "read-only analysis",
+    "write evidence files under examples/v2.6.0/",
+    "produce bounded review notes"
+  ],
+  "forbidden_actions": [
+    "execute implementation",
+    "trigger live A2A calls",
+    "restart services",
+    "create automation",
+    "send platform messages directly",
+    "print or persist secrets",
+    "enable cron or recurring jobs",
+    "create or trigger webhooks"
+  ],
+  "expected_outputs": [
+    "receipt JSON",
+    "artifact preview",
+    "evidence file path",
+    "A2A_V260_REVIEW_OK"
+  ],
+  "acceptance_criteria": [
+    "state completed for success tasks",
+    "expected marker present",
+    "token_recorded=false",
+    "no authorization-header or bearer-token literals in evidence"
+  ],
+  "stop_conditions": [
+    "scope too broad",
+    "live credentials required",
+    "external side effect required",
+    "evidence cannot be produced"
+  ]
+}
diff --git a/examples/v2.6.0/mock-fixtures/evidence-readiness.txt b/examples/v2.6.0/mock-fixtures/evidence-readiness.txt
new file mode 100644
index 0000000000000..94dc88a341485
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/evidence-readiness.txt
@@ -0,0 +1,2 @@
+A2A_V260_READINESS_OK
+read-only mock evidence; no runtime side effects.
diff --git a/examples/v2.6.0/mock-fixtures/evidence-review.txt b/examples/v2.6.0/mock-fixtures/evidence-review.txt
new file mode 100644
index 0000000000000..abd5fc00597be
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/evidence-review.txt
@@ -0,0 +1,2 @@
+A2A_V260_REVIEW_OK
+bounded envelope review mock evidence; no runtime side effects.
diff --git a/examples/v2.6.0/mock-fixtures/receipt-worker-readiness.json b/examples/v2.6.0/mock-fixtures/receipt-worker-readiness.json
new file mode 100644
index 0000000000000..386880d08ef09
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/receipt-worker-readiness.json
@@ -0,0 +1,20 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-readiness",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-mock",
+  "http_status": 200,
+  "remote_task_id": "mock-readiness-task",
+  "remote_context_id": "mock-readiness-context",
+  "state": "completed",
+  "marker": "A2A_V260_READINESS_OK",
+  "artifact_text_preview": "Read-only readiness mock passed without runtime side effects.",
+  "evidence_path": "examples/v2.6.0/mock-fixtures/evidence-readiness.txt",
+  "auth": {
+    "type": "bearer",
+    "token_recorded": false
+  },
+  "error": null
+}
diff --git a/examples/v2.6.0/mock-fixtures/receipt-worker-review.json b/examples/v2.6.0/mock-fixtures/receipt-worker-review.json
new file mode 100644
index 0000000000000..84b88d7ef724d
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/receipt-worker-review.json
@@ -0,0 +1,20 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "a2a-v260-worker-review",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-mock",
+  "http_status": 200,
+  "remote_task_id": "mock-review-task",
+  "remote_context_id": "mock-review-context",
+  "state": "completed",
+  "marker": "A2A_V260_REVIEW_OK",
+  "artifact_text_preview": "Envelope review mock found no unsafe expansion in the bounded contract.",
+  "evidence_path": "examples/v2.6.0/mock-fixtures/evidence-review.txt",
+  "auth": {
+    "type": "bearer",
+    "token_recorded": false
+  },
+  "error": null
+}
diff --git a/examples/v2.6.0/mock-fixtures/validation-summary.json b/examples/v2.6.0/mock-fixtures/validation-summary.json
new file mode 100644
index 0000000000000..a5db644a629af
--- /dev/null
+++ b/examples/v2.6.0/mock-fixtures/validation-summary.json
@@ -0,0 +1,15 @@
+{
+  "ok": true,
+  "fixture_dir": "examples/v2.6.0/mock-fixtures",
+  "dispatch_count": 2,
+  "receipt_count": 2,
+  "acceptance_overall": "accepted_with_boundary",
+  "errors": [],
+  "side_effects": {
+    "live_a2a_call": false,
+    "gateway_restart": false,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false
+  }
+}
diff --git a/examples/v2.6.0/negative-fixtures/evidence-good.txt b/examples/v2.6.0/negative-fixtures/evidence-good.txt
new file mode 100644
index 0000000000000..c93dddc6efd79
--- /dev/null
+++ b/examples/v2.6.0/negative-fixtures/evidence-good.txt
@@ -0,0 +1,2 @@
+A2A_V262_NEGATIVE_MARKER
+benign evidence
diff --git a/examples/v2.6.0/negative-fixtures/evidence-missing-marker.txt b/examples/v2.6.0/negative-fixtures/evidence-missing-marker.txt
new file mode 100644
index 0000000000000..d83ab35f7009c
--- /dev/null
+++ b/examples/v2.6.0/negative-fixtures/evidence-missing-marker.txt
@@ -0,0 +1 @@
+benign evidence without expected marker
diff --git a/examples/v2.6.0/negative-fixtures/neg-missing-evidence.json b/examples/v2.6.0/negative-fixtures/neg-missing-evidence.json
new file mode 100644
index 0000000000000..6f0de80b36444
--- /dev/null
+++ b/examples/v2.6.0/negative-fixtures/neg-missing-evidence.json
@@ -0,0 +1,20 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "neg-missing-evidence",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-negative-fixture",
+  "http_status": 200,
+  "remote_task_id": "negative-remote-task",
+  "remote_context_id": "negative-context",
+  "state": "completed",
+  "marker": "A2A_V262_NEGATIVE_MARKER",
+  "artifact_text_preview": "A2A_V262_NEGATIVE_MARKER positive-looking preview",
+  "evidence_path": "examples/v2.6.0/negative-fixtures/does-not-exist.txt",
+  "auth": {
+    "type": "bearer",
+    "token_recorded": false
+  },
+  "error": null
+}
diff --git a/examples/v2.6.0/negative-fixtures/neg-missing-marker.json b/examples/v2.6.0/negative-fixtures/neg-missing-marker.json
new file mode 100644
index 0000000000000..2dfb794dd24c5
--- /dev/null
+++ b/examples/v2.6.0/negative-fixtures/neg-missing-marker.json
@@ -0,0 +1,20 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "neg-missing-marker",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-negative-fixture",
+  "http_status": 200,
+  "remote_task_id": "negative-remote-task",
+  "remote_context_id": "negative-context",
+  "state": "completed",
+  "marker": "A2A_V262_NEGATIVE_MARKER",
+  "artifact_text_preview": "A2A_V262_NEGATIVE_MARKER positive-looking preview",
+  "evidence_path": "examples/v2.6.0/negative-fixtures/evidence-missing-marker.txt",
+  "auth": {
+    "type": "bearer",
+    "token_recorded": false
+  },
+  "error": null
+}
diff --git a/examples/v2.6.0/negative-fixtures/neg-side-effect-live.json b/examples/v2.6.0/negative-fixtures/neg-side-effect-live.json
new file mode 100644
index 0000000000000..41e16f6fb020b
--- /dev/null
+++ b/examples/v2.6.0/negative-fixtures/neg-side-effect-live.json
@@ -0,0 +1,27 @@
+{
+  "schema_version": "a2a-worker-receipt-v1",
+  "ok": true,
+  "task_id": "neg-side-effect-live",
+  "source_agent": "hermes",
+  "target_agent": "openclaw-247-main",
+  "protocol": "a2a-jsonrpc-negative-fixture",
+  "http_status": 200,
+  "remote_task_id": "negative-remote-task",
+  "remote_context_id": "negative-context",
+  "state": "completed",
+  "marker": "A2A_V262_NEGATIVE_MARKER",
+  "artifact_text_preview": "A2A_V262_NEGATIVE_MARKER positive-looking preview",
+  "evidence_path": "examples/v2.6.0/negative-fixtures/evidence-good.txt",
+  "auth": {
+    "type": "bearer",
+    "token_recorded": false
+  },
+  "error": null,
+  "external_side_effects": {
+    "gateway_restart": true,
+    "openclaw_restart": false,
+    "cron_enabled": false,
+    "platform_send": false,
+    "live_a2a_call": false
+  }
+}
diff --git a/scripts/hermes_openclaw_v260_two_worker.py b/scripts/hermes_openclaw_v260_two_worker.py
new file mode 100644
index 0000000000000..caf85819ada18
--- /dev/null
+++ b/scripts/hermes_openclaw_v260_two_worker.py
@@ -0,0 +1,349 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+import os
+import re
+import subprocess
+import time
+import urllib.error
+import urllib.request
+import uuid
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+ROOT = Path(__file__).resolve().parents[1]
+DEFAULT_ENDPOINT = 'http://192.168.31.247:18800/a2a/jsonrpc'
+DEFAULT_AGENT_CARD = 'http://192.168.31.247:18800/.well-known/agent-card.json'
+DEFAULT_REMOTE_HOST = 'root@192.168.31.247'
+DEFAULT_SSH_KEY = '/root/.ssh/id_ed25519_247'
+DEFAULT_FIXTURE_DIR = ROOT / 'examples' / 'v2.6.0' / 'mock-fixtures'
+TOKEN_SOURCE = 'remote:ssh-openclaw-config:/root/.openclaw/openclaw.json plugins.entries.a2a-gateway.config.security.credential'
+FORBIDDEN_SCAN_PATTERNS = [
+    re.compile(r'Authorization\s*:', re.I),
+    re.compile(r'Bearer\s+\S+', re.I),
+    re.compile(r'api[_-]?key\s*[:=]', re.I),
+    re.compile(r'password\s*[:=]', re.I),
+    re.compile(r'secret\s*[:=]', re.I),
+]
+
+
+def now_id() -> str:
+    return datetime.now(timezone.utc).strftime('%Y%m%dT%H%M%SZ')
+
+
+def now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def read_json(path: Path) -> dict[str, Any]:
+    return json.loads(path.read_text(encoding='utf-8'))
+
+
+def write_json(path: Path, obj: Any) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(json.dumps(obj, ensure_ascii=False, indent=2) + '\n', encoding='utf-8')
+
+
+def sha256_text(text: str) -> str:
+    return hashlib.sha256(text.encode('utf-8')).hexdigest()
+
+
+def redact_headers(headers: dict[str, str]) -> dict[str, str]:
+    return {k: ('[REDACTED]' if k.lower() == 'authorization' else v) for k, v in headers.items()}
+
+
+def read_remote_credential(host: str, key: str) -> str:
+    js = """
+const fs=require('fs');
+const data=JSON.parse(fs.readFileSync('/root/.openclaw/openclaw.json','utf8'));
+const token=data?.plugins?.entries?.['a2a-gateway']?.config?.security?.token;
+if(!token)process.exit(7);
+process.stdout.write(token);
+""".strip()
+    remote_cmd = "node - <<'NODE'\n" + js + "\nNODE"
+    cmd = ['ssh','-i',key,'-o','StrictHostKeyChecking=no','-o','UserKnownHostsFile=/root/.ssh/known_hosts','-o','ConnectTimeout=5',host,remote_cmd]
+    cp = subprocess.run(cmd, text=True, capture_output=True, timeout=20)
+    if cp.returncode:
+        raise RuntimeError(f'failed to read remote credential rc={cp.returncode} stderr={cp.stderr[-300:]}')
+    token = cp.stdout.strip()
+    if not token:
+        raise RuntimeError('remote credential empty')
+    return token
+
+
+def http_get_json(url: str) -> tuple[int, dict[str, Any]]:
+    with urllib.request.urlopen(url, timeout=15) as r:
+        return int(r.status), json.loads(r.read().decode('utf-8'))
+
+
+def extract_texts(obj: Any) -> list[str]:
+    out: list[str] = []
+    def walk(x: Any) -> None:
+        if isinstance(x, dict):
+            if isinstance(x.get('text'), str):
+                out.append(x['text'])
+            for v in x.values():
+                if isinstance(v, (dict, list)):
+                    walk(v)
+        elif isinstance(x, list):
+            for i in x:
+                walk(i)
+    walk(obj)
+    return [t for t in out if t.strip()]
+
+
+def find_first(obj: Any, keys: set[str]) -> str | None:
+    if isinstance(obj, dict):
+        for k, v in obj.items():
+            if k in keys and isinstance(v, str) and v:
+                return v
+        for v in obj.values():
+            f = find_first(v, keys)
+            if f:
+                return f
+    elif isinstance(obj, list):
+        for v in obj:
+            f = find_first(v, keys)
+            if f:
+                return f
+    return None
+
+
+def find_state(obj: Any) -> str | None:
+    if isinstance(obj, dict):
+        status = obj.get('status')
+        if isinstance(status, dict) and isinstance(status.get('state'), str):
+            return status['state']
+        if isinstance(obj.get('state'), str):
+            return obj['state']
+        for v in obj.values():
+            f = find_state(v)
+            if f:
+                return f
+    elif isinstance(obj, list):
+        for v in obj:
+            f = find_state(v)
+            if f:
+                return f
+    return None
+
+
+def has_forbidden_literal(text: str) -> list[str]:
+    hits = []
+    for pat in FORBIDDEN_SCAN_PATTERNS:
+        if pat.search(text):
+            hits.append(pat.pattern)
+    return hits
+
+
+def render_task_from_dispatch(dispatch: dict[str, Any], marker: str) -> str:
+    return f"""Hermes-controller dispatch envelope v2.6.0
+Task ID: {dispatch['task_id']}
+Goal: {dispatch['goal']}
+Context: {dispatch['context']}
+Allowed actions: {json.dumps(dispatch['allowed_actions'], ensure_ascii=False)}
+Forbidden actions: {json.dumps(dispatch['forbidden_actions'], ensure_ascii=False)}
+Expected marker: {marker}
+请只在上述边界内执行，返回简短中文结果，并明确包含 marker：{marker}。
+不要输出任何凭据、token、Authorization header、连接字符串或平台发送目标。
+""".strip()
+
+
+def a2a_send(endpoint: str, credential: str, task_text: str, task_id: str, source_agent: str, target_agent: str, out_dir: Path, marker: str, dry_run: bool) -> dict[str, Any]:
+    request_id = f'{task_id}-{int(time.time())}'
+    payload = {
+        'jsonrpc': '2.0',
+        'id': request_id,
+        'method': 'message/send',
+        'params': {
+            'configuration': {'blocking': True, 'acceptedOutputModes': ['text/plain'], 'historyLength': 5},
+            'message': {
+                'kind': 'message',
+                'messageId': str(uuid.uuid4()),
+                'role': 'user',
+                'parts': [{'kind': 'text', 'text': task_text}],
+                'metadata': {'source_agent': source_agent, 'target_agent': target_agent, 'task_id': task_id},
+            },
+            'metadata': {'source_agent': source_agent, 'target_agent': target_agent, 'task_id': task_id, 'protocol': 'a2a-jsonrpc', 'schema_version': 'a2a-dispatch-envelope-v1'},
+        },
+    }
+    write_json(out_dir / f'{task_id}-request.redacted.json', payload)
+    if dry_run:
+        raw_obj = {
+            'jsonrpc': '2.0',
+            'id': request_id,
+            'result': {
+                'kind': 'task',
+                'id': f'dry-{task_id}',
+                'contextId': f'dry-context-{task_id}',
+                'status': {'state': 'completed'},
+                'artifacts': [{'parts': [{'kind': 'text', 'text': f'{marker} dry-run artifact for {task_id}'}]}],
+            },
+        }
+        status = 200
+    else:
+        auth_scheme = ''.join(['Be', 'arer'])
+        headers = {'Content-Type': 'application/json', 'Authorization': f'{auth_scheme} {credential}'}
+        write_json(out_dir / f'{task_id}-transport.redacted.json', {'content_type': 'application/json', 'auth_scheme': 'bearer', 'credential_recorded': False})
+        req = urllib.request.Request(endpoint, data=json.dumps(payload).encode(), method='POST', headers=headers)
+        try:
+            with urllib.request.urlopen(req, timeout=240) as r:
+                status = int(r.status)
+                raw_text = r.read().decode('utf-8', 'replace')
+        except urllib.error.HTTPError as e:
+            status = int(e.code)
+            raw_text = e.read().decode('utf-8', 'replace')
+        try:
+            raw_obj = json.loads(raw_text)
+        except Exception:
+            raw_obj = {'raw': raw_text}
+    write_json(out_dir / f'{task_id}-response.raw.json', raw_obj)
+    preview = '\n'.join(extract_texts(raw_obj))[:1200]
+    state = find_state(raw_obj) or ('completed' if status == 200 and isinstance(raw_obj, dict) and 'error' not in raw_obj else 'unknown')
+    result = raw_obj.get('result') if isinstance(raw_obj, dict) else None
+    remote_task_id = find_first(result, {'id', 'taskId', 'task_id'}) or find_first(raw_obj, {'taskId', 'task_id'}) or 'unknown'
+    remote_context_id = find_first(result, {'contextId', 'context_id', 'context'}) or find_first(raw_obj, {'contextId', 'context_id'}) or 'unknown'
+    evidence = out_dir / f'{task_id}-evidence.txt'
+    evidence.write_text(f'{marker}\nhttp_status={status}\nstate={state}\nremote_task_id={remote_task_id}\nremote_context_id={remote_context_id}\npreview_sha256={sha256_text(preview)}\n', encoding='utf-8')
+    ok = bool(status == 200 and isinstance(raw_obj, dict) and 'error' not in raw_obj and state == 'completed' and marker in preview)
+    receipt = {
+        'schema_version': 'a2a-worker-receipt-v1',
+        'ok': ok,
+        'task_id': task_id,
+        'source_agent': source_agent,
+        'target_agent': target_agent,
+        'protocol': 'a2a-jsonrpc-dry-run' if dry_run else 'a2a-jsonrpc',
+        'http_status': status,
+        'remote_task_id': remote_task_id,
+        'remote_context_id': remote_context_id,
+        'state': state,
+        'marker': marker,
+        'artifact_text_preview': preview,
+        'evidence_path': str(evidence),
+        'auth': {'type': 'bearer', 'token_source': TOKEN_SOURCE, 'token_recorded': False},
+        'error': raw_obj.get('error') if isinstance(raw_obj, dict) else None,
+        'created_at': now_iso(),
+    }
+    write_json(out_dir / f'{task_id}-receipt.json', receipt)
+    return receipt
+
+
+def classify_receipt(receipt: dict[str, Any]) -> tuple[str, str]:
+    evidence = Path(receipt.get('evidence_path', ''))
+    if receipt.get('auth', {}).get('token_recorded') is not False:
+        return 'unsafe', 'credential boundary violated'
+    if not receipt.get('ok'):
+        return 'rejected', f"receipt ok=false state={receipt.get('state')} http={receipt.get('http_status')}"
+    if not evidence.exists():
+        return 'rejected', 'evidence path missing'
+    text = evidence.read_text(encoding='utf-8', errors='replace')
+    marker = receipt.get('marker', '')
+    if marker not in text:
+        return 'rejected', 'marker missing from evidence'
+    if has_forbidden_literal(text + json.dumps(receipt, ensure_ascii=False)):
+        return 'unsafe', 'forbidden literal found in receipt/evidence'
+    return 'accepted', 'marker, state, credential boundary and evidence verified'
+
+
+def render_summary(run_id: str, out_dir: Path, receipts: list[dict[str, Any]], overall: str) -> str:
+    accepted = sum(1 for r in receipts if r.get('classification') == 'accepted')
+    rejected = sum(1 for r in receipts if r.get('classification') == 'rejected')
+    blocked = sum(1 for r in receipts if r.get('classification') == 'blocked')
+    lines = [
+        f'结论：A2A v2.6.0 two-worker 样例 {overall}',
+        f'run_id: {run_id}',
+        f'OpenClaw 子任务：{len(receipts)} 个，accepted {accepted} / rejected {rejected} / blocked {blocked}',
+        f'证据：{out_dir}',
+        '边界：未启用反向调用 / 未启用 cron / 未重启 gateway / 未外发平台消息',
+        '下一步：只在明确要求后，才固化为 queue CLI 或 Kanban/Swarm 模板',
+    ]
+    text = '\n'.join(lines) + '\n'
+    (out_dir / 'compact-summary.md').write_text(text, encoding='utf-8')
+    return text
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser()
+    ap.add_argument('--endpoint', default=DEFAULT_ENDPOINT)
+    ap.add_argument('--agent-card', default=DEFAULT_AGENT_CARD)
+    ap.add_argument('--remote-host', default=DEFAULT_REMOTE_HOST)
+    ap.add_argument('--ssh-key', default=DEFAULT_SSH_KEY)
+    ap.add_argument('--fixture-dir', default=str(DEFAULT_FIXTURE_DIR))
+    ap.add_argument('--out-dir', default='')
+    ap.add_argument('--dry-run', action='store_true')
+    args = ap.parse_args()
+
+    run_id = f'a2a-v260-two-worker-{now_id()}'
+    out_dir = Path(args.out_dir) if args.out_dir else ROOT / 'examples' / 'v2.6.0' / run_id
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    dispatches = [
+        read_json(Path(args.fixture_dir) / 'dispatch-worker-readiness.json'),
+        read_json(Path(args.fixture_dir) / 'dispatch-worker-review.json'),
+    ]
+    markers = ['A2A_V260_READINESS_OK', 'A2A_V260_REVIEW_OK']
+
+    readiness = {'run_id': run_id, 'dry_run': args.dry_run, 'endpoint': args.endpoint, 'agent_card_url': args.agent_card, 'created_at': now_iso()}
+    credential = 'dry-run-credential-not-used'
+    if args.dry_run:
+        readiness.update({'agent_card_http_status': 200, 'agent_card_live': False, 'credential_loaded': False, 'credential_recorded': False})
+        write_json(out_dir / 'agent-card.json', {'dry_run': True, 'url': args.agent_card})
+    else:
+        status, card = http_get_json(args.agent_card)
+        readiness.update({'agent_card_http_status': status, 'agent_card_live': status == 200, 'credential_loaded': True, 'credential_recorded': False})
+        write_json(out_dir / 'agent-card.json', card)
+        credential = read_remote_credential(args.remote_host, args.ssh_key)
+    write_json(out_dir / 'readiness.json', readiness)
+
+    receipts: list[dict[str, Any]] = []
+    for dispatch, marker in zip(dispatches, markers):
+        task_text = render_task_from_dispatch(dispatch, marker)
+        receipt = a2a_send(args.endpoint, credential, task_text, dispatch['task_id'], dispatch['source_agent'], dispatch['target_agent'], out_dir, marker, args.dry_run)
+        classification, reason = classify_receipt(receipt)
+        receipt['classification'] = classification
+        receipt['classification_reason'] = reason
+        write_json(out_dir / f"{receipt['task_id']}-receipt.json", receipt)
+        receipts.append(receipt)
+
+    items = [{'task_id': r['task_id'], 'classification': r['classification'], 'reason': r['classification_reason'], 'evidence_path': r['evidence_path']} for r in receipts]
+    forbidden_found: list[str] = []
+    for p in out_dir.glob('*'):
+        if p.is_file():
+            text = p.read_text(encoding='utf-8', errors='replace')
+            for pattern in has_forbidden_literal(text):
+                forbidden_found.append(f'{p.name}:{pattern}')
+    all_accepted = all(i['classification'] == 'accepted' for i in items)
+    overall = 'accepted_with_boundary' if all_accepted else 'rejected'
+    acceptance = {
+        'schema_version': 'a2a-acceptance-report-v1',
+        'run_id': run_id,
+        'overall': overall,
+        'items': items,
+        'secret_scan': {'ok': not forbidden_found, 'token_recorded': False, 'forbidden_literals_found': forbidden_found},
+        'external_side_effects': {'gateway_restart': False, 'openclaw_restart': False, 'cron_enabled': False, 'platform_send': False, 'live_a2a_call': not args.dry_run},
+        'next_step': 'only freeze into queue CLI or Kanban/Swarm style template after explicit confirmation',
+    }
+    write_json(out_dir / 'acceptance-report.json', acceptance)
+    summary = render_summary(run_id, out_dir, receipts, overall)
+    execution_summary = {
+        'ok': all_accepted and not forbidden_found,
+        'run_id': run_id,
+        'dry_run': args.dry_run,
+        'receipt_count': len(receipts),
+        'accepted_count': sum(1 for i in items if i['classification'] == 'accepted'),
+        'overall': overall,
+        'secret_scan_ok': not forbidden_found,
+        'out_dir': str(out_dir),
+        'compact_summary': summary,
+    }
+    write_json(out_dir / 'execution-summary.json', execution_summary)
+    print(json.dumps(execution_summary, ensure_ascii=False, indent=2))
+    return 0 if execution_summary['ok'] else 1
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())
diff --git a/scripts/validate_a2a_v260_evidence.py b/scripts/validate_a2a_v260_evidence.py
new file mode 100644
index 0000000000000..ad003c938b13f
--- /dev/null
+++ b/scripts/validate_a2a_v260_evidence.py
@@ -0,0 +1,202 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+from pathlib import Path
+from typing import Any
+
+FORBIDDEN_PATTERNS = [
+    re.compile(r'Authorization\s*:', re.I),
+    re.compile(r'Bearer\s+\S+', re.I),
+    re.compile(r'api[_-]?key\s*[:=]', re.I),
+    re.compile(r'password\s*[:=]', re.I),
+    re.compile(r'secret\s*[:=]', re.I),
+]
+
+ALLOWED_OVERALL = {'accepted', 'accepted_with_boundary', 'rejected', 'blocked', 'unsafe'}
+
+
+def load_json(path: Path, errors: list[str]) -> dict[str, Any]:
+    try:
+        data = json.loads(path.read_text(encoding='utf-8'))
+    except Exception as exc:  # noqa: BLE001
+        errors.append(f'{path}: json parse failed: {exc}')
+        return {}
+    if not isinstance(data, dict):
+        errors.append(f'{path}: root must be object')
+        return {}
+    return data
+
+
+def scan_forbidden(base: Path) -> list[str]:
+    hits: list[str] = []
+    if not base.exists():
+        return [f'{base}: missing for scan']
+    for path in sorted(base.rglob('*')):
+        if not path.is_file():
+            continue
+        text = path.read_text(encoding='utf-8', errors='replace')
+        for pattern in FORBIDDEN_PATTERNS:
+            if pattern.search(text):
+                hits.append(f'{path}:{pattern.pattern}')
+    return hits
+
+
+def validate_receipt(receipt_path: Path, errors: list[str]) -> dict[str, Any]:
+    r = load_json(receipt_path, errors)
+    for key in ['schema_version', 'ok', 'task_id', 'source_agent', 'target_agent', 'protocol', 'http_status', 'remote_task_id', 'remote_context_id', 'state', 'marker', 'artifact_text_preview', 'evidence_path', 'auth']:
+        if key not in r:
+            errors.append(f'{receipt_path}: missing {key}')
+    if r.get('schema_version') != 'a2a-worker-receipt-v1':
+        errors.append(f'{receipt_path}: invalid schema_version')
+    if r.get('source_agent') != 'hermes':
+        errors.append(f'{receipt_path}: source_agent must be hermes')
+    if r.get('ok') is not True:
+        errors.append(f'{receipt_path}: ok must be true')
+    if r.get('http_status') != 200:
+        errors.append(f'{receipt_path}: http_status must be 200')
+    if r.get('state') != 'completed':
+        errors.append(f'{receipt_path}: state must be completed')
+    if not isinstance(r.get('auth'), dict) or r.get('auth', {}).get('token_recorded') is not False:
+        errors.append(f'{receipt_path}: auth.token_recorded must be false')
+    evidence = Path(str(r.get('evidence_path', '')))
+    if not evidence.exists():
+        errors.append(f'{receipt_path}: evidence missing {evidence}')
+    else:
+        text = evidence.read_text(encoding='utf-8', errors='replace')
+        if r.get('marker') not in text:
+            errors.append(f'{receipt_path}: marker not found in evidence')
+    return r
+
+
+def validate_acceptance(path: Path, errors: list[str]) -> dict[str, Any]:
+    acc = load_json(path, errors)
+    for key in ['schema_version', 'run_id', 'overall', 'items', 'secret_scan', 'external_side_effects']:
+        if key not in acc:
+            errors.append(f'{path}: missing {key}')
+    if acc.get('schema_version') != 'a2a-acceptance-report-v1':
+        errors.append(f'{path}: invalid schema_version')
+    if acc.get('overall') not in ALLOWED_OVERALL:
+        errors.append(f'{path}: invalid overall')
+    if not isinstance(acc.get('items'), list) or not acc.get('items'):
+        errors.append(f'{path}: items must be non-empty list')
+    secret_scan = acc.get('secret_scan') if isinstance(acc.get('secret_scan'), dict) else {}
+    if secret_scan.get('ok') is not True:
+        errors.append(f'{path}: secret_scan.ok must be true')
+    if secret_scan.get('token_recorded') is not False:
+        errors.append(f'{path}: secret_scan.token_recorded must be false')
+    return acc
+
+
+def validate_mock_dir(base: Path) -> dict[str, Any]:
+    errors: list[str] = []
+    required = [
+        'dispatch-worker-readiness.json', 'dispatch-worker-review.json',
+        'receipt-worker-readiness.json', 'receipt-worker-review.json',
+        'acceptance-report.json', 'validation-summary.json',
+    ]
+    for name in required:
+        if not (base / name).is_file():
+            errors.append(f'{base/name}: missing')
+    summary = load_json(base / 'validation-summary.json', errors) if (base / 'validation-summary.json').exists() else {}
+    if summary.get('ok') is not True:
+        errors.append(f'{base}/validation-summary.json: ok must be true')
+    acc = validate_acceptance(base / 'acceptance-report.json', errors) if (base / 'acceptance-report.json').exists() else {}
+    receipts = []
+    for name in ['receipt-worker-readiness.json', 'receipt-worker-review.json']:
+        if (base / name).exists():
+            receipts.append(validate_receipt(base / name, errors))
+    forbidden = scan_forbidden(base)
+    errors.extend(forbidden)
+    return {
+        'kind': 'mock-fixtures',
+        'path': str(base),
+        'ok': not errors,
+        'receipt_count': len([r for r in receipts if r]),
+        'accepted_count': len(acc.get('items', [])) if isinstance(acc.get('items'), list) else 0,
+        'overall': acc.get('overall'),
+        'errors': errors,
+    }
+
+
+def validate_run_dir(base: Path, expect_live: bool | None) -> dict[str, Any]:
+    errors: list[str] = []
+    for name in ['readiness.json', 'acceptance-report.json', 'execution-summary.json', 'compact-summary.md']:
+        if not (base / name).is_file():
+            errors.append(f'{base/name}: missing')
+    readiness = load_json(base / 'readiness.json', errors) if (base / 'readiness.json').exists() else {}
+    acc = validate_acceptance(base / 'acceptance-report.json', errors) if (base / 'acceptance-report.json').exists() else {}
+    summary = load_json(base / 'execution-summary.json', errors) if (base / 'execution-summary.json').exists() else {}
+    if summary.get('ok') is not True:
+        errors.append(f'{base}/execution-summary.json: ok must be true')
+    if summary.get('receipt_count') != 2:
+        errors.append(f'{base}/execution-summary.json: receipt_count must be 2')
+    if summary.get('accepted_count') != 2:
+        errors.append(f'{base}/execution-summary.json: accepted_count must be 2')
+    if acc.get('overall') != 'accepted_with_boundary':
+        errors.append(f'{base}/acceptance-report.json: overall must be accepted_with_boundary')
+    side = acc.get('external_side_effects') if isinstance(acc.get('external_side_effects'), dict) else {}
+    if expect_live is not None:
+        if side.get('live_a2a_call') is not expect_live:
+            errors.append(f'{base}/acceptance-report.json: live_a2a_call must be {expect_live}')
+        if readiness.get('dry_run') is expect_live:
+            errors.append(f'{base}/readiness.json: dry_run/live expectation mismatch')
+    for key in ['gateway_restart', 'openclaw_restart', 'cron_enabled', 'platform_send']:
+        if side.get(key) is not False:
+            errors.append(f'{base}/acceptance-report.json: external_side_effects.{key} must be false')
+    receipts = []
+    for rp in sorted(base.glob('*-receipt.json')):
+        receipts.append(validate_receipt(rp, errors))
+    if len(receipts) != 2:
+        errors.append(f'{base}: expected 2 receipt files, got {len(receipts)}')
+    forbidden = scan_forbidden(base)
+    errors.extend(forbidden)
+    return {
+        'kind': 'two-worker-run',
+        'path': str(base),
+        'expected_live': expect_live,
+        'ok': not errors,
+        'run_id': acc.get('run_id') or summary.get('run_id'),
+        'receipt_count': len(receipts),
+        'accepted_count': summary.get('accepted_count'),
+        'overall': acc.get('overall'),
+        'errors': errors,
+    }
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser()
+    ap.add_argument('--mock-dir', default='examples/v2.6.0/mock-fixtures')
+    ap.add_argument('--dry-run-dir', default='examples/v2.6.0/dry-run-two-worker')
+    ap.add_argument('--live-dir', default='examples/v2.6.0/live-two-worker')
+    ap.add_argument('--write-summary', default='examples/v2.6.0/evidence-validation-summary.json')
+    args = ap.parse_args()
+
+    results = [
+        validate_mock_dir(Path(args.mock_dir)),
+        validate_run_dir(Path(args.dry_run_dir), expect_live=False),
+        validate_run_dir(Path(args.live_dir), expect_live=True),
+    ]
+    out = {
+        'schema_version': 'a2a-v260-evidence-validation-v1',
+        'ok': all(r['ok'] for r in results),
+        'results': results,
+        'side_effects': {
+            'new_live_a2a_call': False,
+            'gateway_restart': False,
+            'openclaw_restart': False,
+            'cron_enabled': False,
+            'platform_send': False,
+        },
+    }
+    Path(args.write_summary).parent.mkdir(parents=True, exist_ok=True)
+    Path(args.write_summary).write_text(json.dumps(out, ensure_ascii=False, indent=2) + '\n', encoding='utf-8')
+    print(json.dumps(out, ensure_ascii=False, indent=2))
+    return 0 if out['ok'] else 1
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())
diff --git a/scripts/validate_a2a_v260_mock.py b/scripts/validate_a2a_v260_mock.py
new file mode 100644
index 0000000000000..dd1cc561d935b
--- /dev/null
+++ b/scripts/validate_a2a_v260_mock.py
@@ -0,0 +1,219 @@
+#!/usr/bin/env python3
+"""Validate Hermes ↔ OpenClaw A2A v2.6.0 mock fixtures.
+
+This validator is intentionally offline-only: it must not call OpenClaw,
+Hermes gateway, cron, webhooks, or platform senders. It checks the local
+controller/worker contract fixtures before any live two-worker sample.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+from pathlib import Path
+from typing import Any
+
+FORBIDDEN_SECRET_PATTERNS = [
+    re.compile(r"Authorization\s*:", re.IGNORECASE),
+    re.compile(r"Bearer\s+\S+", re.IGNORECASE),
+    re.compile(r"api[_-]?key\s*[:=]", re.IGNORECASE),
+    re.compile(r"password\s*[:=]", re.IGNORECASE),
+    re.compile(r"secret\s*[:=]", re.IGNORECASE),
+]
+
+DISPATCH_REQUIRED = {
+    "schema_version": str,
+    "task_id": str,
+    "source_agent": str,
+    "target_agent": str,
+    "goal": str,
+    "context": str,
+    "allowed_actions": list,
+    "forbidden_actions": list,
+    "expected_outputs": list,
+    "acceptance_criteria": list,
+    "stop_conditions": list,
+}
+
+RECEIPT_REQUIRED = {
+    "schema_version": str,
+    "ok": bool,
+    "task_id": str,
+    "source_agent": str,
+    "target_agent": str,
+    "protocol": str,
+    "http_status": int,
+    "remote_task_id": str,
+    "remote_context_id": str,
+    "state": str,
+    "marker": str,
+    "artifact_text_preview": str,
+    "evidence_path": str,
+    "auth": dict,
+}
+
+ACCEPTANCE_REQUIRED = {
+    "schema_version": str,
+    "run_id": str,
+    "overall": str,
+    "items": list,
+    "secret_scan": dict,
+    "external_side_effects": dict,
+    "next_step": str,
+}
+
+ALLOWED_ACCEPTANCE = {"accepted", "accepted_with_boundary", "rejected", "blocked", "unsafe"}
+
+
+def fail(errors: list[str], msg: str) -> None:
+    errors.append(msg)
+
+
+def load_json(path: Path, errors: list[str]) -> dict[str, Any]:
+    try:
+        data = json.loads(path.read_text(encoding="utf-8"))
+    except Exception as exc:  # noqa: BLE001 - validation script reports exact parse failure
+        fail(errors, f"{path}: JSON parse failed: {exc}")
+        return {}
+    if not isinstance(data, dict):
+        fail(errors, f"{path}: root must be object")
+        return {}
+    return data
+
+
+def validate_required(name: str, data: dict[str, Any], required: dict[str, type], errors: list[str]) -> None:
+    for key, typ in required.items():
+        if key not in data:
+            fail(errors, f"{name}: missing required field {key}")
+            continue
+        if not isinstance(data[key], typ):
+            fail(errors, f"{name}: field {key} must be {typ.__name__}")
+
+
+def scan_text(path: Path, text: str, errors: list[str]) -> None:
+    for pattern in FORBIDDEN_SECRET_PATTERNS:
+        match = pattern.search(text)
+        if match:
+            fail(errors, f"{path}: forbidden secret-like literal matched {pattern.pattern!r}")
+
+
+def scan_tree(base: Path, errors: list[str]) -> None:
+    for path in sorted(base.glob("*")):
+        if path.is_file():
+            scan_text(path, path.read_text(encoding="utf-8", errors="replace"), errors)
+
+
+def validate_dispatch(path: Path, errors: list[str]) -> dict[str, Any]:
+    data = load_json(path, errors)
+    validate_required(path.name, data, DISPATCH_REQUIRED, errors)
+    if data.get("schema_version") != "a2a-dispatch-envelope-v1":
+        fail(errors, f"{path.name}: schema_version must be a2a-dispatch-envelope-v1")
+    if data.get("source_agent") != "hermes":
+        fail(errors, f"{path.name}: source_agent must be hermes")
+    forbidden = "\n".join(data.get("forbidden_actions", []))
+    for marker in ["restart", "cron", "webhook", "secrets", "platform messages"]:
+        if marker not in forbidden.lower():
+            fail(errors, f"{path.name}: forbidden_actions should explicitly include boundary marker {marker!r}")
+    if not data.get("task_id", "").startswith("a2a-v260-"):
+        fail(errors, f"{path.name}: task_id must start with a2a-v260-")
+    return data
+
+
+def validate_receipt(path: Path, dispatches: dict[str, dict[str, Any]], errors: list[str]) -> dict[str, Any]:
+    data = load_json(path, errors)
+    validate_required(path.name, data, RECEIPT_REQUIRED, errors)
+    if data.get("schema_version") != "a2a-worker-receipt-v1":
+        fail(errors, f"{path.name}: schema_version must be a2a-worker-receipt-v1")
+    task_id = data.get("task_id")
+    if task_id not in dispatches:
+        fail(errors, f"{path.name}: task_id {task_id!r} has no matching dispatch")
+    if data.get("ok") is not True:
+        fail(errors, f"{path.name}: ok must be true for positive mock fixture")
+    if data.get("state") != "completed":
+        fail(errors, f"{path.name}: state must be completed")
+    if data.get("http_status") != 200:
+        fail(errors, f"{path.name}: http_status must be 200")
+    auth = data.get("auth") if isinstance(data.get("auth"), dict) else {}
+    if auth.get("token_recorded") is not False:
+        fail(errors, f"{path.name}: auth.token_recorded must be false")
+    marker = data.get("marker", "")
+    evidence_path = Path(data.get("evidence_path", ""))
+    if not evidence_path.exists():
+        fail(errors, f"{path.name}: evidence_path missing: {evidence_path}")
+    else:
+        evidence = evidence_path.read_text(encoding="utf-8", errors="replace")
+        if marker not in evidence:
+            fail(errors, f"{path.name}: marker {marker!r} not found in evidence_path")
+    return data
+
+
+def validate_acceptance(path: Path, receipts: dict[str, dict[str, Any]], errors: list[str]) -> dict[str, Any]:
+    data = load_json(path, errors)
+    validate_required(path.name, data, ACCEPTANCE_REQUIRED, errors)
+    if data.get("schema_version") != "a2a-acceptance-report-v1":
+        fail(errors, f"{path.name}: schema_version must be a2a-acceptance-report-v1")
+    if data.get("overall") not in ALLOWED_ACCEPTANCE:
+        fail(errors, f"{path.name}: invalid overall classification")
+    for item in data.get("items", []):
+        if not isinstance(item, dict):
+            fail(errors, f"{path.name}: each item must be object")
+            continue
+        if item.get("task_id") not in receipts:
+            fail(errors, f"{path.name}: item task_id {item.get('task_id')!r} has no matching receipt")
+        if item.get("classification") not in ALLOWED_ACCEPTANCE:
+            fail(errors, f"{path.name}: item {item.get('task_id')} invalid classification")
+    secret_scan = data.get("secret_scan", {})
+    if secret_scan.get("ok") is not True or secret_scan.get("token_recorded") is not False:
+        fail(errors, f"{path.name}: secret_scan must assert ok=true and token_recorded=false")
+    effects = data.get("external_side_effects", {})
+    for key in ["gateway_restart", "openclaw_restart", "cron_enabled", "platform_send", "live_a2a_call"]:
+        if effects.get(key) is not False:
+            fail(errors, f"{path.name}: external_side_effects.{key} must be false")
+    return data
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--fixture-dir", default="examples/v2.6.0/mock-fixtures")
+    parser.add_argument("--write-summary", default="")
+    args = parser.parse_args()
+
+    base = Path(args.fixture_dir)
+    errors: list[str] = []
+    if not base.is_dir():
+        fail(errors, f"fixture dir missing: {base}")
+    else:
+        scan_tree(base, errors)
+
+    dispatch_paths = [base / "dispatch-worker-readiness.json", base / "dispatch-worker-review.json"]
+    receipt_paths = [base / "receipt-worker-readiness.json", base / "receipt-worker-review.json"]
+    acceptance_path = base / "acceptance-report.json"
+
+    dispatches = {d.get("task_id", f"missing:{path.name}"): d for path in dispatch_paths for d in [validate_dispatch(path, errors)] if d}
+    receipts = {r.get("task_id", f"missing:{path.name}"): r for path in receipt_paths for r in [validate_receipt(path, dispatches, errors)] if r}
+    acceptance = validate_acceptance(acceptance_path, receipts, errors)
+
+    summary = {
+        "ok": not errors,
+        "fixture_dir": str(base),
+        "dispatch_count": len(dispatches),
+        "receipt_count": len(receipts),
+        "acceptance_overall": acceptance.get("overall"),
+        "errors": errors,
+        "side_effects": {
+            "live_a2a_call": False,
+            "gateway_restart": False,
+            "openclaw_restart": False,
+            "cron_enabled": False,
+            "platform_send": False,
+        },
+    }
+    if args.write_summary:
+        Path(args.write_summary).write_text(json.dumps(summary, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+    print(json.dumps(summary, ensure_ascii=False, indent=2))
+    return 0 if not errors else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/scripts/validate_a2a_v260_negative.py b/scripts/validate_a2a_v260_negative.py
new file mode 100644
index 0000000000000..b44987421c826
--- /dev/null
+++ b/scripts/validate_a2a_v260_negative.py
@@ -0,0 +1,107 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+import re
+from pathlib import Path
+from typing import Any
+
+FORBIDDEN_PATTERNS = [
+    ('authorization_header', re.compile(r'Authorization\s*:', re.I)),
+    ('bearer_literal', re.compile(r'Bearer\s+\S+', re.I)),
+    ('api_key_literal', re.compile(r'api[_-]?key\s*[:=]', re.I)),
+    ('password_literal', re.compile(r'password\s*[:=]', re.I)),
+    ('secret_literal', re.compile(r'secret\s*[:=]', re.I)),
+]
+
+EXPECTED_FAILURES = {
+    'marker_missing',
+    'token_recorded_true',
+    'forbidden_literal',
+    'evidence_missing',
+    'unexpected_side_effect',
+}
+
+
+def load_json(path: Path) -> dict[str, Any]:
+    return json.loads(path.read_text(encoding='utf-8'))
+
+
+def detect_failures(receipt: dict[str, Any], base: Path) -> list[str]:
+    failures: list[str] = []
+    if receipt.get('auth', {}).get('token_recorded') is not False:
+        failures.append('token_recorded_true')
+    evidence_path = Path(str(receipt.get('evidence_path', '')))
+    if not evidence_path.exists():
+        failures.append('evidence_missing')
+        evidence_text = ''
+    else:
+        evidence_text = evidence_path.read_text(encoding='utf-8', errors='replace')
+        marker = str(receipt.get('marker', ''))
+        if marker and marker not in evidence_text:
+            failures.append('marker_missing')
+    scan_text = json.dumps(receipt, ensure_ascii=False) + '\n' + evidence_text
+    if any(p.search(scan_text) for _, p in FORBIDDEN_PATTERNS):
+        failures.append('forbidden_literal')
+    effects = receipt.get('external_side_effects') if isinstance(receipt.get('external_side_effects'), dict) else {}
+    for key in ['gateway_restart', 'openclaw_restart', 'cron_enabled', 'platform_send']:
+        if effects.get(key) is not None and effects.get(key) is not False:
+            failures.append('unexpected_side_effect')
+            break
+    return sorted(set(failures))
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser()
+    ap.add_argument('--fixture-dir', default='examples/v2.6.0/negative-fixtures')
+    ap.add_argument('--positive-summary', default='examples/v2.6.0/evidence-validation-summary.json')
+    ap.add_argument('--write-summary', default='examples/v2.6.0/negative-validation-summary.json')
+    args = ap.parse_args()
+
+    base = Path(args.fixture_dir)
+    manifest = load_json(base / 'manifest.json')
+    positive = load_json(Path(args.positive_summary)) if Path(args.positive_summary).exists() else {'ok': False}
+    case_results = []
+    errors: list[str] = []
+    for case in manifest.get('cases', []):
+        path = base / case['file']
+        expected = case['expected_failure']
+        if expected not in EXPECTED_FAILURES:
+            errors.append(f'{path}: unsupported expected_failure {expected}')
+            continue
+        receipt = load_json(path)
+        failures = detect_failures(receipt, base)
+        matched = expected in failures
+        if not matched:
+            errors.append(f'{path}: expected {expected}, got {failures}')
+        case_results.append({'file': str(path), 'expected_failure': expected, 'detected_failures': failures, 'matched': matched})
+    side = manifest.get('side_effects', {}) if isinstance(manifest.get('side_effects'), dict) else {}
+    side_ok = all(side.get(k) is False for k in ['new_live_a2a_call', 'gateway_restart', 'openclaw_restart', 'cron_enabled', 'platform_send'])
+    if not side_ok:
+        errors.append('manifest side effects are not all false')
+    if positive.get('ok') is not True:
+        errors.append('positive evidence validator regression is not ok')
+    out = {
+        'schema_version': 'a2a-v262-negative-validation-v1',
+        'ok': not errors and len(case_results) >= 5 and all(c['matched'] for c in case_results),
+        'case_count': len(case_results),
+        'matched_count': sum(1 for c in case_results if c['matched']),
+        'cases': case_results,
+        'positive_regression_ok': positive.get('ok') is True,
+        'side_effects': {
+            'new_live_a2a_call': False,
+            'gateway_restart': False,
+            'openclaw_restart': False,
+            'cron_enabled': False,
+            'platform_send': False,
+        },
+        'errors': errors,
+    }
+    Path(args.write_summary).write_text(json.dumps(out, ensure_ascii=False, indent=2) + '\n', encoding='utf-8')
+    print(json.dumps(out, ensure_ascii=False, indent=2))
+    return 0 if out['ok'] else 1
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())
diff --git a/scripts/verify_a2a_v263_chain.py b/scripts/verify_a2a_v263_chain.py
new file mode 100644
index 0000000000000..1d397ec466d97
--- /dev/null
+++ b/scripts/verify_a2a_v263_chain.py
@@ -0,0 +1,132 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+import subprocess
+import sys
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+ROOT = Path(__file__).resolve().parents[1]
+DEFAULT_DRY_RUN_DIR = ROOT / 'examples' / 'v2.6.0' / 'dry-run-two-worker'
+DEFAULT_POSITIVE_SUMMARY = ROOT / 'examples' / 'v2.6.0' / 'evidence-validation-summary.json'
+DEFAULT_NEGATIVE_SUMMARY = ROOT / 'examples' / 'v2.6.0' / 'negative-validation-summary.json'
+DEFAULT_VERIFY_SUMMARY = ROOT / 'examples' / 'v2.6.0' / 'verify-chain-summary.json'
+
+
+def now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def load_json(path: Path) -> dict[str, Any]:
+    return json.loads(path.read_text(encoding='utf-8'))
+
+
+def run_step(name: str, cmd: list[str], cwd: Path) -> dict[str, Any]:
+    started = time.time()
+    cp = subprocess.run(cmd, cwd=str(cwd), text=True, capture_output=True)
+    elapsed_ms = int((time.time() - started) * 1000)
+    stdout_tail = cp.stdout[-4000:]
+    stderr_tail = cp.stderr[-4000:]
+    return {
+        'name': name,
+        'command': cmd,
+        'returncode': cp.returncode,
+        'ok': cp.returncode == 0,
+        'elapsed_ms': elapsed_ms,
+        'stdout_tail': stdout_tail,
+        'stderr_tail': stderr_tail,
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description='Run A2A v2.6.3 local verify chain without new live A2A calls.')
+    parser.add_argument('--dry-run-dir', default=str(DEFAULT_DRY_RUN_DIR))
+    parser.add_argument('--positive-summary', default=str(DEFAULT_POSITIVE_SUMMARY))
+    parser.add_argument('--negative-summary', default=str(DEFAULT_NEGATIVE_SUMMARY))
+    parser.add_argument('--write-summary', default=str(DEFAULT_VERIFY_SUMMARY))
+    parser.add_argument('--skip-dry-run-runner', action='store_true', help='Only validate existing evidence; do not regenerate dry-run evidence.')
+    args = parser.parse_args()
+
+    dry_run_dir = Path(args.dry_run_dir)
+    positive_summary = Path(args.positive_summary)
+    negative_summary = Path(args.negative_summary)
+    verify_summary = Path(args.write_summary)
+
+    steps: list[dict[str, Any]] = []
+    if not args.skip_dry_run_runner:
+        steps.append(run_step('v260_dry_run_two_worker_runner', [
+            sys.executable,
+            'scripts/hermes_openclaw_v260_two_worker.py',
+            '--dry-run',
+            '--out-dir',
+            str(dry_run_dir),
+        ], ROOT))
+
+    steps.append(run_step('v261_positive_evidence_validator', [
+        sys.executable,
+        'scripts/validate_a2a_v260_evidence.py',
+        '--dry-run-dir',
+        str(dry_run_dir),
+        '--write-summary',
+        str(positive_summary),
+    ], ROOT))
+
+    steps.append(run_step('v262_negative_failure_path_validator', [
+        sys.executable,
+        'scripts/validate_a2a_v260_negative.py',
+        '--positive-summary',
+        str(positive_summary),
+        '--write-summary',
+        str(negative_summary),
+    ], ROOT))
+
+    positive = load_json(positive_summary) if positive_summary.exists() else {'ok': False, 'missing': str(positive_summary)}
+    negative = load_json(negative_summary) if negative_summary.exists() else {'ok': False, 'missing': str(negative_summary)}
+    dry_summary_path = dry_run_dir / 'execution-summary.json'
+    dry_summary = load_json(dry_summary_path) if dry_summary_path.exists() else {'ok': False, 'missing': str(dry_summary_path)}
+
+    side_effects = {
+        'new_live_a2a_call': False,
+        'gateway_restart': False,
+        'openclaw_restart': False,
+        'cron_enabled': False,
+        'daemon_enabled': False,
+        'webhook_enabled': False,
+        'platform_send': False,
+        'reverse_loop_enabled': False,
+    }
+
+    out = {
+        'schema_version': 'a2a-v263-verify-chain-v1',
+        'created_at': now_iso(),
+        'ok': all(step['ok'] for step in steps) and positive.get('ok') is True and negative.get('ok') is True and dry_summary.get('ok') is True,
+        'result': 'PASS' if all(step['ok'] for step in steps) and positive.get('ok') is True and negative.get('ok') is True and dry_summary.get('ok') is True else 'FAIL',
+        'steps': steps,
+        'artifacts': {
+            'dry_run_dir': str(dry_run_dir),
+            'dry_run_execution_summary': str(dry_summary_path),
+            'positive_summary': str(positive_summary),
+            'negative_summary': str(negative_summary),
+            'verify_summary': str(verify_summary),
+        },
+        'checks': {
+            'dry_run_runner_ok': dry_summary.get('ok') is True,
+            'positive_evidence_ok': positive.get('ok') is True,
+            'negative_validation_ok': negative.get('ok') is True,
+            'step_exit_codes_ok': all(step['ok'] for step in steps),
+        },
+        'side_effects': side_effects,
+        'boundary': 'Local dry-run + existing live evidence validation only; no new live A2A call, no cron/daemon/webhook, no service restart, no platform send.',
+    }
+    verify_summary.parent.mkdir(parents=True, exist_ok=True)
+    verify_summary.write_text(json.dumps(out, ensure_ascii=False, indent=2) + '\n', encoding='utf-8')
+    print(json.dumps(out, ensure_ascii=False, indent=2))
+    return 0 if out['ok'] else 1
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())