Files
awoooi/docs/evaluations/ai_agent_12_agent_war_room_2026-06-14.json

342 lines
14 KiB
JSON
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
{
"schema_version": "ai_agent_12_agent_war_room_v1",
"generated_at": "2026-06-14T14:20:00+08:00",
"program_status": {
"current_priority": "P1",
"current_task_id": "P2-142",
"next_task_id": "P2-143",
"overall_completion_percent": 72,
"read_only_mode": true,
"runtime_authority": "12_agent_war_room_read_only_no_live_write",
"status_note": "12 個邏輯工位與 12 份只讀審查已收斂;本 snapshot 只把協作狀態產品化,不開 runtime writer、Telegram send、Bot API、SDK 安裝、付費 API 或 production write。"
},
"source_refs": [
"docs/ai/AI_AGENT_12_AGENT_WAR_ROOM_2026-06-14.md",
"docs/ai/AI_AGENT_AUTOMATION_WORKLIST_2026-06-04.md",
"docs/HARD_RULES.md",
"docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md",
"docs/12-agent-game-rules.md",
"https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b/modelcard",
"https://github.com/openai/openai-agents-python/releases",
"https://github.com/langchain-ai/langgraph/releases",
"https://adk.dev/2.0/"
],
"coordination_model": {
"logical_agent_count": 12,
"subagent_batch_limit": 6,
"subagent_batch_count": 2,
"coordination_mode": "read_only_evidence_then_gate",
"arbiter": "openclaw",
"memory_owner": "hermes",
"replay_owner": "nemotron",
"gateway_owner": "telegram_ops_liaison",
"summary": "12 個工位分批收集證據OpenClaw 只做仲裁與 HITL gateHermes 做記憶與 RAGNemoTron 做離線回放與候選評估Telegram Ops 做低噪音報告與 receipt。"
},
"agent_roles": [
{
"agent_id": "agent_01_openclaw_arbiter",
"display_name": "OpenClaw 仲裁者",
"primary_agent": "OpenClaw",
"war_room_role": "仲裁與風險裁決",
"review_status": "read_only_review_completed",
"risk_tier": "critical",
"work_units": 7,
"evidence_items": 6,
"blocker_count": 4,
"approval_required_count": 5,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "維護 action permission matrix並把其他 Agent 的證據轉成 gate verdict。"
},
{
"agent_id": "agent_02_hermes_rag",
"display_name": "Hermes 記憶/RAG",
"primary_agent": "Hermes",
"war_room_role": "記憶、RAG、報告與知識治理",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 6,
"evidence_items": 6,
"blocker_count": 3,
"approval_required_count": 4,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "把 War Room 結果轉成 RAG chunk、報告摘要與 owner review 草稿,但不保存未公開協作內容。"
},
{
"agent_id": "agent_03_nemotron_replay",
"display_name": "NemoTron 回放專家",
"primary_agent": "NemoTron",
"war_room_role": "離線回放、模型比較與執行建議",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 6,
"evidence_items": 5,
"blocker_count": 5,
"approval_required_count": 6,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "先走 source refresh 與 5-record smoke 批准包,不進 full replay、shadow/canary 或 production route。"
},
{
"agent_id": "agent_04_sre_sentinel",
"display_name": "SRE Sentinel",
"primary_agent": "SRE",
"war_room_role": "可靠性、SLO、告警與觀測契約",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 8,
"evidence_items": 8,
"blocker_count": 5,
"approval_required_count": 4,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "把 service green、DR blocked、governance debt 分開報告;降噪只產 proposal不 reload 或 silence。"
},
{
"agent_id": "agent_05_security_sentinel",
"display_name": "Security Sentinel",
"primary_agent": "Security",
"war_room_role": "IwoooS、redaction、source-control 與 runtime gate",
"review_status": "read_only_review_completed",
"risk_tier": "critical",
"work_units": 7,
"evidence_items": 7,
"blocker_count": 5,
"approval_required_count": 6,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "維持只讀 evidence、owner response envelope、public bundle、source-control、runtime 五層 gate。"
},
{
"agent_id": "agent_06_devops_commander",
"display_name": "DevOps Commander",
"primary_agent": "DevOps",
"war_room_role": "Gitea CD、ArgoCD、K8s 與 release gate",
"review_status": "read_only_review_completed",
"risk_tier": "critical",
"work_units": 8,
"evidence_items": 9,
"blocker_count": 6,
"approval_required_count": 7,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "把 NetworkPolicy、migration、runner、NEXT_PUBLIC 與 ArgoCD Degraded 漂移列入批准包。"
},
{
"agent_id": "agent_07_data_dr_guardian",
"display_name": "Data / DR Guardian",
"primary_agent": "Data DR",
"war_room_role": "備份、異地、escrow、restore drill 與 cold-start gate",
"review_status": "read_only_review_completed",
"risk_tier": "critical",
"work_units": 7,
"evidence_items": 7,
"blocker_count": 5,
"approval_required_count": 6,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "核心備份偏綠但 DR 不得宣告完成;先補 5 個 non-secret escrow evidence marker。"
},
{
"agent_id": "agent_08_supply_chain_scout",
"display_name": "Supply Chain Scout",
"primary_agent": "Supply Chain",
"war_room_role": "套件、Docker、CVE、license、版本漂移與升級批准包",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 7,
"evidence_items": 7,
"blocker_count": 5,
"approval_required_count": 7,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "只讀檢查 manifest / lockfile / Docker surface外部 CVE、套件升級與 image pull 需批准。"
},
{
"agent_id": "agent_09_product_ui_curator",
"display_name": "Product UI Curator",
"primary_agent": "Product UI",
"war_room_role": "治理 UI、動畫、redaction 與 mobile smoke",
"review_status": "read_only_review_completed",
"risk_tier": "medium",
"work_units": 6,
"evidence_items": 6,
"blocker_count": 3,
"approval_required_count": 2,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "把 12-Agent 狀態做成可掃描、mobile 不爆版、不可誤讀為授權的只讀 UI。"
},
{
"agent_id": "agent_10_qa_verifier",
"display_name": "QA Verifier",
"primary_agent": "QA",
"war_room_role": "pytest、typecheck、build、production smoke、console、overflow 與禁用文字掃描",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 7,
"evidence_items": 8,
"blocker_count": 4,
"approval_required_count": 3,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "每輪交付輸出 PASS / CONDITIONAL PASS / BLOCKED並附命令、route、viewport、commit 或 deploy marker。"
},
{
"agent_id": "agent_11_market_scout",
"display_name": "Agent Market Scout",
"primary_agent": "Market Scout",
"war_room_role": "AI Agent 市場雷達與候選批准包",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 6,
"evidence_items": 8,
"blocker_count": 4,
"approval_required_count": 6,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "刷新 Agent Market snapshot將新版本與候選轉成 OpenClaw 可仲裁、NemoTron 可回放的批准包。"
},
{
"agent_id": "agent_12_telegram_ops_liaison",
"display_name": "Telegram Ops Liaison",
"primary_agent": "Telegram Ops",
"war_room_role": "日報、週報、月報、receipt、低噪音通知與審核卡片",
"review_status": "read_only_review_completed",
"risk_tier": "high",
"work_units": 7,
"evidence_items": 7,
"blocker_count": 5,
"approval_required_count": 5,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"next_action": "建立統一 report receipt低風險進摘要高風險只產審核入口不替代 runtime authorization。"
}
],
"workload_summary": {
"total_work_units": 82,
"total_evidence_items": 84,
"status_report_visible": true,
"chart_ready": true,
"daily_report_required": true,
"weekly_report_required": true,
"monthly_report_required": true,
"standard_agent_metrics": [
"agent_id",
"agent_role",
"task_count",
"operation_count",
"success_count",
"failed_count",
"auto_resolved_count",
"human_review_count",
"approval_required_count",
"mcp_call_count",
"mcp_blocked_count",
"avg_latency_ms",
"tokens_in",
"tokens_out",
"estimated_cost",
"telegram_sent_count",
"dedup_suppressed_count",
"receipt_missing_count",
"km_entries_created",
"playbook_updates",
"recommendations_created"
]
},
"risk_policy": {
"low_risk_auto_handle": "文件、只讀報告、UI 顯示、測試報告可自動處理並摘要回報。",
"medium_risk_auto_handle": "read-only、已授權、可回滾、Gateway scope 通過者可自動處理;設定、外部服務、成本或狀態不確定則升級審核。",
"high_risk_requires_approval": "production write、kubectl apply、ArgoCD sync、Telegram 實發、付費 API、SDK 安裝、host update、DB migration、restore、rollback、NetworkPolicy、workflow 修改均需統帥批准。",
"openclaw_gate_required": true,
"qa_verifier_required": true,
"security_gate_required": true
},
"reporting_contract": {
"daily": {
"required": true,
"delivery_mode": "summary_only_unless_failed_or_blocked",
"sections": ["總告警數", "自動處理量", "人工審核量", "KM 新增", "PlayBook 更新", "成本與 token", "Telegram receipt"]
},
"weekly": {
"required": true,
"delivery_mode": "operator_reviewed_digest",
"sections": ["Agent 工作量", "成功率", "部署數", "critical alert", "resolved rate", "K3s / pod / HPA", "AI 成本"]
},
"monthly": {
"required": true,
"delivery_mode": "governance_chart_and_approval_backlog",
"sections": ["每 Agent 任務量", "自動化節省時間", "高風險案例", "市場評估", "成本趨勢", "RAG / KM 成長", "Telegram 送達率"]
}
},
"market_watch_contract": {
"cadence": "每週一 09:00 台北時間;重大版本、新框架或高訊號 benchmark 觸發臨時複核。",
"auto_collect_allowed": ["官方 release metadata", "版本號", "content hash", "source failure", "公開 benchmark 摘要", "既有內部 replay / smoke 結果"],
"approval_required": ["SDK 安裝", "付費 API / NIM 呼叫", "外部帳號登入", "full replay", "shadow / canary", "production route", "provider switch", "lockfile write", "workflow 修改", "Telegram 實發"],
"p0_refresh_candidates": [
"Nemotron 3 Ultra 550B-A55B",
"OpenAI Agents SDK 最新 release",
"LangGraph 1.2.5",
"Google ADK 2.0",
"Microsoft Agent Framework 1.0"
],
"candidate_package_required_fields": ["官方來源", "版本差異", "建議角色", "market score", "AWOOOI replay/smoke 計畫", "成本與資料邊界", "SDK/依賴影響", "Telegram/告警策略", "rollback", "測試清單", "OpenClaw 仲裁", "HITL 到期時間"]
},
"telegram_contract": {
"direct_send_allowed": false,
"bot_api_call_allowed": false,
"success_immediate_send_allowed": false,
"action_required_digest_allowed_after_approval": true,
"failure_only_escalation": true,
"dedup_required": true,
"receipt_required": true,
"forbidden_actions": ["logOut", "close", "setWebhook", "deleteWebhook", "direct_bot_api_send", "unsafe_callback_execution"],
"summary": "Agent 12 統一處理 Telegram Gateway、Channel Hub、report receipt、去重、脫敏與 action-required digest高風險 Telegram 卡片只可作審核入口,不可取代 runtime authorization。"
},
"display_redaction_contract": {
"redaction_required": true,
"conversation_transcript_display_allowed": false,
"raw_prompt_display_allowed": false,
"private_reasoning_display_allowed": false,
"secret_value_display_allowed": false,
"raw_runtime_payload_display_allowed": false,
"frontend_display_policy": "前端只顯示角色、狀態、工作量、阻擋項、批准邊界與脫敏 evidence refs禁止顯示未公開協作逐字稿、prompt、session、browser context、secret、token 或 callback 原文。"
},
"rollups": {
"agent_role_count": 12,
"read_only_review_completed_count": 12,
"subagent_batch_limit": 6,
"subagent_batch_count": 2,
"approval_required_total": 61,
"blocker_total": 54,
"total_work_units": 82,
"total_evidence_items": 84,
"daily_report_required_count": 1,
"weekly_report_required_count": 1,
"monthly_report_required_count": 1,
"market_refresh_candidate_count": 5,
"live_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"production_write_count": 0,
"paid_api_call_count": 0,
"sdk_install_count": 0,
"secret_read_count": 0,
"destructive_operation_count": 0
}
}