wooo/awoooi
1
1
Fork 0
Code Issues 2 Pull Requests 1 Actions 4 Packages Projects Releases Wiki Activity

feat(agents): expose controlled executor handoff runway
Some checks failed
Code Review / ai-code-review (push) Successful in 22s
Details
CD Pipeline / tests (push) Successful in 1m47s
Details
CD Pipeline / build-and-deploy (push) Successful in 6m20s
Details
CD Pipeline / post-deploy-checks (push) Successful in 2m18s
Details
Ansible / Reboot Recovery Contract / validate (push) Has been cancelled
Details

Browse Source
This commit is contained in:
Your Name
2026-06-27 11:42:21 +08:00
parent fccd8874fc
commit b2b51ecbf2
11 changed files with 2092 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
docs/LOGBOOK.md
View File

@@ -1,3 +1,28 @@
## 2026-06-27｜P2-415 AI Agent 受控 Executor 交接跑道：API / 前台 / 測試完成
**背景**：P2-409 已把 high 風險從「人工 owner review」調整為可走 controlled apply queue，critical 仍保留 break-glass 邊界；本段承接 P2-409 / P2-410 / P2-411，補上可被產品與正式 API 讀回的受控 executor handoff 跑道，避免只停留在 UI 文案或口頭批准。
**完成內容**：
- 新增 `docs/evaluations/ai_agent_controlled_executor_handoff_2026-06-27.json` 與 schema；明確列出 7 個 handoff packets，其中 `5` 個 high 風險可交給受控 executor，`2` 個 critical 維持 break-glass only。
- 新增 `ai_agent_controlled_executor_handoff.py` loader；強制驗證 allowlist、Ansible check-mode、rollback、post-action verifier、Telegram evidence、KM writeback、PlayBook trust writeback，並強制 dispatch / live apply / Telegram send / Bot API / KM write / production write / secret read / host write / kubectl / destructive count 全部為 `0`。
- 新增 `GET /api/v1/agents/agent-controlled-executor-handoff`；端點只讀 committed snapshot，不 dispatch executor、不寫 Gateway queue、不送 Telegram、不呼叫 Bot API、不讀 secret、不改主機。
- `/zh-TW/governance?tab=automation-inventory` 新增 P2-415「受控 Executor 交接跑道」卡片，呈現 handoff packets、executor routes、verifier / learning binding、critical break-glass 與正式寫入總數。
- `ApiClient`、繁中 / 英文 message catalog、錯誤 gate 與載入 gate 已同步，P2-410 / P2-411 失敗也納入 automation inventory 錯誤判定。
**驗證結果**：
- `python3.11 -m json.tool`：P2-415 snapshot / schema、`zh-TW.json`、`en.json` 均通過。
- `python3.11 -m py_compile apps/api/src/services/ai_agent_controlled_executor_handoff.py apps/api/src/api/v1/agents.py`：通過。
- `pytest` P2-415 專項：`9 passed`。
- P2-409 / P2-410 / P2-411 / P2-415 回歸：`26 passed`。
- `pnpm --filter @awoooi/web typecheck`：通過。
- `git diff --check`：通過。
**目前真相邊界**：
- P2-415 功能完成度：`100%`（source / API / 前台 / tests）。
- High 風險 handoff ready：`5/5`；critical break-glass：`2/2`。
- controlled executor dispatch / live apply / Gateway queue write / Telegram send / Bot API / KM write / PlayBook trust write / production write / secret read / paid API / host write / kubectl / destructive operation：全部 `0`。
- 本段沒有正式下發 Ansible apply、沒有 Telegram live send、沒有主機寫入、沒有 secret value collection、沒有 destructive operation。
## 2026-06-27｜D1J 修復候選升級合約：受控自動化閉環可視化
**背景**：Telegram / AwoooP 告警卡已能指出 `repair_candidate_draft_ready_owner_review`，但 Work Items / Approvals 仍主要呈現「需人工」與長文字，操作員看不到 AI 已準備好的無寫入演練、Verifier 與 KM / PlayBook / Script 資產沉澱槽位，容易被誤判為 AI Agent 沒有任何自動化進展。

617
docs/evaluations/ai_agent_controlled_executor_handoff_2026-06-27.json Normal file
View File

@@ -0,0 +1,617 @@
{
"schema_version": "ai_agent_controlled_executor_handoff_v1",
"generated_at": "2026-06-27T01:20:00+08:00",
"program_status": {
"overall_completion_percent": 100,
"current_priority": "P0",
"current_task_id": "P2-415",
"next_task_id": "P2-416",
"read_only_mode": true,
"runtime_authority": "controlled_executor_handoff_readback_no_live_apply",
"status_note": "P2-415 承接 P2-409 controlled apply queue，把 high 風險候選整理成可交給 executor 的 handoff packet：allowlist、Ansible check-mode、rollback、post-action verifier、Telegram evidence、KM / PlayBook trust 回寫全部要可讀。此 readback 不直接執行 live apply。"
},
"source_refs": [
"docs/evaluations/ai_agent_high_risk_owner_review_queue_2026-06-19.json",
"docs/evaluations/ai_agent_action_audit_ledger_2026-06-19.json",
"docs/evaluations/ai_agent_action_owner_acceptance_event_bus_2026-06-19.json",
"docs/evaluations/ai_agent_report_runtime_readiness_2026-06-12.json",
"docs/evaluations/ai_agent_runtime_write_gate_review_2026-06-12.json",
"docs/evaluations/ai_agent_post_write_verifier_package_2026-06-12.json",
"docs/evaluations/ai_agent_learning_writeback_approval_package_2026-06-11.json",
"docs/evaluations/ai_agent_telegram_receipt_approval_package_2026-06-11.json"
],
"source_readbacks": [
{
"readback_id": "p2_409_controlled_apply_queue",
"source_schema_version": "ai_agent_high_risk_owner_review_queue_v1",
"source_ref": "docs/evaluations/ai_agent_high_risk_owner_review_queue_2026-06-19.json",
"endpoint": "GET /api/v1/agents/agent-high-risk-owner-review-queue",
"owner_agent": "openclaw",
"status": "loaded",
"key_readback": "high 風險已轉 controlled_apply_queue；critical / secret / destructive / paid / force-push 維持 break-glass。",
"next_action": "將 5 個 high packet 映射到 executor handoff route。"
},
{
"readback_id": "p2_410_action_audit_ledger",
"source_schema_version": "ai_agent_action_audit_ledger_v1",
"source_ref": "docs/evaluations/ai_agent_action_audit_ledger_2026-06-19.json",
"endpoint": "GET /api/v1/agents/agent-action-audit-ledger",
"owner_agent": "hermes",
"status": "loaded",
"key_readback": "審計事件模板、redacted evidence refs、verifier receipt gate 已可讀。",
"next_action": "讓 executor handoff packet 帶入 immutable audit fields。"
},
{
"readback_id": "p2_411_handoff_event_bus",
"source_schema_version": "ai_agent_action_owner_acceptance_event_bus_v1",
"source_ref": "docs/evaluations/ai_agent_action_owner_acceptance_event_bus_2026-06-19.json",
"endpoint": "GET /api/v1/agents/agent-action-owner-acceptance-event-bus",
"owner_agent": "hermes",
"status": "loaded",
"key_readback": "交接事件、RAG proposal 與 verifier gate 已建好，但舊語意仍偏 no-write。",
"next_action": "把 high 風險 handoff 從 owner hold 改成 controlled executor runway。"
},
{
"readback_id": "runtime_readiness_low_medium_high",
"source_schema_version": "ai_agent_report_runtime_readiness_v1",
"source_ref": "docs/evaluations/ai_agent_report_runtime_readiness_2026-06-12.json",
"endpoint": "GET /api/v1/agents/agent-report-runtime-readiness",
"owner_agent": "openclaw",
"status": "loaded",
"key_readback": "low / medium / high policy 已允許 auto after guard，critical 才需要 break-glass。",
"next_action": "把 policy 轉成 executor handoff allowlist 與 post-action verifier binding。"
},
{
"readback_id": "runtime_write_gate_review",
"source_schema_version": "ai_agent_runtime_write_gate_review_v1",
"source_ref": "docs/evaluations/ai_agent_runtime_write_gate_review_2026-06-12.json",
"endpoint": "GET /api/v1/agents/agent-runtime-write-gate-review",
"owner_agent": "sre",
"status": "loaded",
"key_readback": "runtime write gate 已定義 dry-run hash、post-write verifier、redaction 欄位。",
"next_action": "高風險 handoff packet 必須引用 check-mode 與 post-write verifier ref。"
},
{
"readback_id": "post_write_verifier_package",
"source_schema_version": "ai_agent_post_write_verifier_package_v1",
"source_ref": "docs/evaluations/ai_agent_post_write_verifier_package_2026-06-12.json",
"endpoint": "GET /api/v1/agents/agent-post-write-verifier-package",
"owner_agent": "nemotron",
"status": "loaded",
"key_readback": "post-write verifier package、rollback lane 與 failure lane 已可讀。",
"next_action": "每個 controlled executor packet 必須綁定 verifier 與 rollback lane。"
},
{
"readback_id": "learning_writeback_package",
"source_schema_version": "ai_agent_learning_writeback_approval_package_v1",
"source_ref": "docs/evaluations/ai_agent_learning_writeback_approval_package_2026-06-11.json",
"endpoint": "GET /api/v1/agents/agent-learning-writeback-approval-package",
"owner_agent": "hermes",
"status": "loaded",
"key_readback": "KM、timeline learning、PlayBook trust 與 replay score 回寫欄位已定義。",
"next_action": "讓 executor handoff packet 產出可回寫的 learning receipt preview。"
},
{
"readback_id": "telegram_receipt_package",
"source_schema_version": "ai_agent_telegram_receipt_approval_package_v1",
"source_ref": "docs/evaluations/ai_agent_telegram_receipt_approval_package_2026-06-11.json",
"endpoint": "GET /api/v1/agents/agent-telegram-receipt-approval-package",
"owner_agent": "hermes",
"status": "loaded",
"key_readback": "Telegram receipt、queue、delivery、ack、failure、retry 欄位已定義；不得包含 token 或原始 chat id。",
"next_action": "handoff 成功 / verifier 失敗 / rollback queued 都要能生成脫敏 Telegram evidence。"
}
],
"handoff_truth": {
"p2_409_controlled_apply_queue_loaded": true,
"p2_410_audit_ledger_loaded": true,
"p2_411_handoff_event_bus_loaded": true,
"runtime_readiness_loaded": true,
"runtime_write_gate_loaded": true,
"post_write_verifier_loaded": true,
"learning_writeback_loaded": true,
"telegram_receipt_loaded": true,
"high_risk_controlled_executor_handoff_ready": true,
"high_risk_owner_review_required": false,
"critical_break_glass_required": true,
"allowlist_route_required": true,
"ansible_check_mode_required": true,
"rollback_plan_required": true,
"post_action_verifier_required": true,
"telegram_evidence_required": true,
"km_writeback_required": true,
"playbook_trust_writeback_required": true,
"controlled_executor_dispatch_enabled": false,
"live_apply_enabled": false,
"critical_auto_bypass_allowed": false,
"gateway_queue_write_enabled": false,
"telegram_send_enabled": false,
"bot_api_call_enabled": false,
"km_write_enabled": false,
"playbook_trust_write_enabled": false,
"production_write_enabled": false,
"secret_read_enabled": false,
"paid_api_call_enabled": false,
"host_write_enabled": false,
"kubectl_action_enabled": false,
"destructive_operation_enabled": false,
"controlled_executor_dispatch_count_24h": 0,
"live_apply_count_24h": 0,
"gateway_queue_write_count_24h": 0,
"telegram_send_count_24h": 0,
"bot_api_call_count_24h": 0,
"km_write_count_24h": 0,
"playbook_trust_write_count_24h": 0,
"production_write_count_24h": 0,
"secret_read_count_24h": 0,
"paid_api_call_count_24h": 0,
"host_write_count_24h": 0,
"kubectl_action_count_24h": 0,
"destructive_operation_count_24h": 0,
"truth_note": "high 風險不再停在人工佇列；5 個 high packet 已具備 controlled executor handoff 條件。此端點只讀回 handoff runway，實際 dispatch / live apply / Telegram send / KM writeback 仍由 executor 與 verifier 計數回報。"
},
"executor_handoff_packets": [
{
"packet_id": "handoff_high_security_response",
"source_queue_item_id": "high_security_response_queue",
"display_name": "資安回應受控 executor 交接",
"risk_tier": "high",
"owner_agent": "openclaw",
"executor_agent": "security",
"executor_type": "ansible_playbook",
"handoff_status": "ready_for_controlled_executor",
"controlled_route_id": "allowlisted_security_response_controlled_apply",
"playbook_ref": "infra/ansible/playbooks/security-controlled-response.yml",
"mcp_tool_ref": "mcp://security/readiness-and-diff",
"check_mode_ref": "ansible-check/security-controlled-response",
"verifier_ref": "verifier://security-post-action-readback",
"rollback_ref": "rollback://security-no-secret-restore-plan",
"telegram_evidence_ref": "telegram-evidence://security-controlled-apply-redacted",
"km_writeback_ref": "km://security-controlled-apply-learning",
"playbook_trust_ref": "playbook-trust://security-controlled-response",
"allowlist_match": true,
"check_mode_passed": true,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": false,
"break_glass_required": false,
"controlled_executor_handoff_allowed": true,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["secret read", "credentialed exploit", "active response without verifier"],
"next_gate": "dispatch worker 從此 packet 取 check-mode receipt、執行 controlled apply，並把 verifier 結果回寫 KM / PlayBook trust。"
},
{
"packet_id": "handoff_high_data_config_apply",
"source_queue_item_id": "high_data_config_apply_queue",
"display_name": "資料與設定受控 executor 交接",
"risk_tier": "high",
"owner_agent": "sre",
"executor_agent": "devops",
"executor_type": "ansible_playbook",
"handoff_status": "ready_for_controlled_executor",
"controlled_route_id": "allowlisted_config_drift_controlled_apply",
"playbook_ref": "infra/ansible/playbooks/config-drift-controlled-apply.yml",
"mcp_tool_ref": "mcp://config/rendered-diff",
"check_mode_ref": "ansible-check/config-drift-controlled-apply",
"verifier_ref": "verifier://config-route-smoke",
"rollback_ref": "rollback://config-source-of-truth-revert",
"telegram_evidence_ref": "telegram-evidence://config-apply-redacted",
"km_writeback_ref": "km://config-drift-learning",
"playbook_trust_ref": "playbook-trust://config-controlled-apply",
"allowlist_match": true,
"check_mode_passed": true,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": false,
"break_glass_required": false,
"controlled_executor_handoff_allowed": true,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["DB DROP", "restore apply", "retention prune"],
"next_gate": "dispatch worker 只能對 source-of-truth diff 執行 check-mode 通過的 controlled apply。"
},
{
"packet_id": "handoff_high_live_telegram_gateway_send",
"source_queue_item_id": "high_live_telegram_gateway_send_queue",
"display_name": "Telegram Gateway 受控 executor 交接",
"risk_tier": "high",
"owner_agent": "hermes",
"executor_agent": "hermes",
"executor_type": "telegram_gateway_queue",
"handoff_status": "ready_for_controlled_executor",
"controlled_route_id": "allowlisted_failure_only_telegram_gateway",
"playbook_ref": "infra/ansible/playbooks/telegram-gateway-route-check.yml",
"mcp_tool_ref": "mcp://telegram-gateway/no-secret-preview",
"check_mode_ref": "gateway-check/failure-only-dedupe-preview",
"verifier_ref": "verifier://telegram-receipt-redacted-readback",
"rollback_ref": "rollback://telegram-dedupe-and-silence-revert",
"telegram_evidence_ref": "telegram-evidence://gateway-message-shape-redacted",
"km_writeback_ref": "km://telegram-delivery-learning",
"playbook_trust_ref": "playbook-trust://telegram-gateway-controlled-send",
"allowlist_match": true,
"check_mode_passed": true,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": false,
"break_glass_required": false,
"controlled_executor_handoff_allowed": true,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["direct Bot API", "token read", "raw chat id display"],
"next_gate": "dispatch worker 必須走 Gateway、dedupe key 與 redacted receipt；不得直接 Bot API。"
},
{
"packet_id": "handoff_high_report_source_gap_work_item_write",
"source_queue_item_id": "high_report_source_gap_work_item_write_queue",
"display_name": "報表缺口與 KM 回寫 executor 交接",
"risk_tier": "high",
"owner_agent": "hermes",
"executor_agent": "nemotron",
"executor_type": "km_playbook_writer",
"handoff_status": "ready_for_controlled_executor",
"controlled_route_id": "allowlisted_report_gap_learning_writeback",
"playbook_ref": "playbooks/report-source-gap-learning.yml",
"mcp_tool_ref": "mcp://knowledge/redacted-learning-packet",
"check_mode_ref": "writer-check/report-gap-learning-preview",
"verifier_ref": "verifier://km-playbook-trust-receipt",
"rollback_ref": "rollback://km-learning-entry-revert-preview",
"telegram_evidence_ref": "telegram-evidence://learning-writeback-summary",
"km_writeback_ref": "km://report-source-gap-learning",
"playbook_trust_ref": "playbook-trust://report-gap-remediation",
"allowlist_match": true,
"check_mode_passed": true,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": false,
"break_glass_required": false,
"controlled_executor_handoff_allowed": true,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["raw report payload write", "private reasoning write", "unbounded embedding write"],
"next_gate": "dispatch worker 只寫 redacted learning packet，並以 verifier receipt 更新 trust delta。"
},
{
"packet_id": "handoff_high_host_kubectl_orchestrated_change",
"source_queue_item_id": "high_host_kubectl_orchestrated_change_queue",
"display_name": "主機與 K8s 受控 executor 交接",
"risk_tier": "high",
"owner_agent": "sre",
"executor_agent": "sre",
"executor_type": "ansible_playbook",
"handoff_status": "ready_for_controlled_executor",
"controlled_route_id": "allowlisted_host_k8s_check_mode_apply",
"playbook_ref": "infra/ansible/playbooks/host-k8s-controlled-apply.yml",
"mcp_tool_ref": "mcp://sre/topology-and-health-readback",
"check_mode_ref": "ansible-check/host-k8s-controlled-apply",
"verifier_ref": "verifier://host-k8s-health-postcheck",
"rollback_ref": "rollback://host-k8s-controlled-revert",
"telegram_evidence_ref": "telegram-evidence://host-k8s-apply-summary",
"km_writeback_ref": "km://host-k8s-remediation-learning",
"playbook_trust_ref": "playbook-trust://host-k8s-controlled-apply",
"allowlist_match": true,
"check_mode_passed": true,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": false,
"break_glass_required": false,
"controlled_executor_handoff_allowed": true,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["reboot", "node drain", "force rollout without verifier"],
"next_gate": "dispatch worker 必須先完成 target selector、check-mode、blast-radius guard 與 rollback stop condition。"
},
{
"packet_id": "handoff_critical_model_cost_provider_change",
"source_queue_item_id": "critical_model_cost_provider_change_queue",
"display_name": "模型角色與費用 break-glass",
"risk_tier": "critical",
"owner_agent": "openclaw",
"executor_agent": "openclaw",
"executor_type": "break_glass_only",
"handoff_status": "critical_break_glass_only",
"controlled_route_id": "blocked_critical_model_cost_provider_boundary",
"playbook_ref": "adr://market-replay-shadow-canary-required",
"mcp_tool_ref": "mcp://agent-market/scorecard-readback",
"check_mode_ref": "not-applicable-critical-break-glass",
"verifier_ref": "verifier://agent-market-replay-shadow-canary",
"rollback_ref": "rollback://provider-route-fallback",
"telegram_evidence_ref": "telegram-evidence://critical-cost-provider-summary",
"km_writeback_ref": "km://agent-market-decision-learning",
"playbook_trust_ref": "playbook-trust://agent-provider-role-decision",
"allowlist_match": false,
"check_mode_passed": false,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": true,
"break_glass_required": true,
"controlled_executor_handoff_allowed": false,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["OpenClaw replacement", "paid provider switch", "cost quota change"],
"next_gate": "必須先有市場分數、replay、shadow、canary 與費用邊界，不能由一般 high 風險自動化覆蓋。"
},
{
"packet_id": "handoff_critical_secret_paid_provider_boundary",
"source_queue_item_id": "critical_secret_paid_provider_boundary_queue",
"display_name": "secret 與付費 provider break-glass",
"risk_tier": "critical",
"owner_agent": "security",
"executor_agent": "security",
"executor_type": "break_glass_only",
"handoff_status": "critical_break_glass_only",
"controlled_route_id": "blocked_critical_secret_paid_provider_boundary",
"playbook_ref": "policy://secret-paid-provider-break-glass",
"mcp_tool_ref": "mcp://security/secret-metadata-only",
"check_mode_ref": "not-applicable-critical-break-glass",
"verifier_ref": "verifier://secret-boundary-and-cost-cap",
"rollback_ref": "rollback://provider-secret-metadata-revert",
"telegram_evidence_ref": "telegram-evidence://critical-secret-boundary-summary",
"km_writeback_ref": "km://secret-boundary-learning",
"playbook_trust_ref": "playbook-trust://secret-provider-boundary",
"allowlist_match": false,
"check_mode_passed": false,
"rollback_plan_ready": true,
"post_action_verifier_ready": true,
"telegram_evidence_ready": true,
"km_writeback_ready": true,
"playbook_trust_writeback_ready": true,
"owner_response_required": true,
"break_glass_required": true,
"controlled_executor_handoff_allowed": false,
"live_apply_performed": false,
"side_effect_count": 0,
"blocked_runtime_actions": ["secret value read", "paid API expansion", "privacy egress change"],
"next_gate": "只允許 metadata 與 evidence ref；secret value、付費 provider 擴張與隱私外送必須 break-glass。"
}
],
"executor_routes": [
{
"route_id": "ansible_check_mode_controlled_apply",
"display_name": "Ansible check-mode controlled apply",
"executor_agent": "sre",
"route_status": "ready_for_handoff",
"required_inputs": ["target selector", "source-of-truth ref", "check-mode receipt", "rollback owner", "post-action verifier"],
"blocked_actions": ["reboot", "node drain", "destructive DB operation"],
"live_apply_allowed_by_this_readback": false
},
{
"route_id": "mcp_tool_registry_preflight",
"display_name": "MCP tool registry preflight",
"executor_agent": "openclaw",
"route_status": "ready_for_handoff",
"required_inputs": ["tool scope", "risk tier", "allowed action", "blocked action", "redacted evidence ref"],
"blocked_actions": ["unregistered tool call", "raw secret volume access"],
"live_apply_allowed_by_this_readback": false
},
{
"route_id": "telegram_gateway_redacted_evidence",
"display_name": "Telegram Gateway redacted evidence",
"executor_agent": "hermes",
"route_status": "ready_for_handoff",
"required_inputs": ["canonical room env", "dedupe key", "message shape", "receipt expectation", "redaction proof"],
"blocked_actions": ["direct Bot API", "raw chat id display", "token read"],
"live_apply_allowed_by_this_readback": false
},
{
"route_id": "km_playbook_trust_writer",
"display_name": "KM / PlayBook trust writer",
"executor_agent": "nemotron",
"route_status": "ready_for_handoff",
"required_inputs": ["redacted learning packet", "matched playbook id", "verifier receipt", "rollback criteria", "trust delta"],
"blocked_actions": ["private reasoning write", "unbounded embedding write"],
"live_apply_allowed_by_this_readback": false
},
{
"route_id": "post_action_verifier_and_rollback",
"display_name": "Post-action verifier and rollback lane",
"executor_agent": "sre",
"route_status": "ready_for_handoff",
"required_inputs": ["pre-state ref", "post-state ref", "failure threshold", "rollback stop condition"],
"blocked_actions": ["verifier without baseline", "rollback without stop condition"],
"live_apply_allowed_by_this_readback": false
}
],
"verifier_bindings": [
{
"binding_id": "binding_ansible_check_mode",
"display_name": "Ansible check-mode receipt binding",
"owner_agent": "sre",
"required_before_dispatch": true,
"ready_count": 5,
"blocked_count": 0,
"failure_if_missing": "缺 check-mode receipt 時不得 dispatch controlled executor。"
},
{
"binding_id": "binding_rollback_owner",
"display_name": "Rollback owner and stop condition binding",
"owner_agent": "sre",
"required_before_dispatch": true,
"ready_count": 5,
"blocked_count": 0,
"failure_if_missing": "缺 rollback owner 或 stop condition 時不得 apply。"
},
{
"binding_id": "binding_post_action_verifier",
"display_name": "Post-action verifier binding",
"owner_agent": "nemotron",
"required_before_dispatch": true,
"ready_count": 5,
"blocked_count": 0,
"failure_if_missing": "缺 verifier ref 時不得視為自動化閉環。"
},
{
"binding_id": "binding_learning_writeback",
"display_name": "KM / PlayBook trust writeback binding",
"owner_agent": "hermes",
"required_before_dispatch": true,
"ready_count": 5,
"blocked_count": 0,
"failure_if_missing": "缺 learning receipt 時不得更新完成度。"
},
{
"binding_id": "binding_telegram_evidence",
"display_name": "Telegram redacted evidence binding",
"owner_agent": "hermes",
"required_before_dispatch": true,
"ready_count": 5,
"blocked_count": 0,
"failure_if_missing": "缺 redacted Telegram evidence 時不得對外宣稱已處理。"
}
],
"learning_writeback_contracts": [
{
"contract_id": "km_execution_receipt",
"display_name": "KM execution receipt",
"owner_agent": "hermes",
"target_store": "knowledge_entries",
"writeback_status": "ready_for_executor_receipt",
"required_fields": ["decision id", "executor route", "verifier result", "redacted evidence refs", "rollback outcome"],
"runtime_write_performed": false
},
{
"contract_id": "playbook_trust_delta",
"display_name": "PlayBook trust delta",
"owner_agent": "openclaw",
"target_store": "playbooks",
"writeback_status": "ready_for_executor_receipt",
"required_fields": ["matched playbook id", "success or failure", "verifier confidence", "negative reinforcement reason"],
"runtime_write_performed": false
},
{
"contract_id": "timeline_event_append",
"display_name": "Timeline event append",
"owner_agent": "hermes",
"target_store": "timeline_events",
"writeback_status": "ready_for_executor_receipt",
"required_fields": ["agent role", "affected scope", "decision reason", "executor status", "post-check result"],
"runtime_write_performed": false
}
],
"activation_boundaries": {
"committed_snapshot_read_allowed": true,
"controlled_executor_handoff_preview_allowed": true,
"ansible_check_mode_receipt_preview_allowed": true,
"mcp_tool_registry_route_preview_allowed": true,
"post_action_verifier_binding_preview_allowed": true,
"telegram_evidence_preview_allowed": true,
"km_playbook_trust_writeback_preview_allowed": true,
"controlled_executor_dispatch_enabled": false,
"live_apply_enabled": false,
"gateway_queue_write_enabled": false,
"telegram_send_enabled": false,
"bot_api_call_enabled": false,
"km_write_enabled": false,
"playbook_trust_write_enabled": false,
"production_write_enabled": false,
"secret_read_enabled": false,
"paid_api_call_enabled": false,
"host_write_enabled": false,
"kubectl_action_enabled": false,
"destructive_operation_enabled": false
},
"display_redaction_contract": {
"redaction_required": true,
"raw_tool_output_display_allowed": false,
"raw_runtime_payload_display_allowed": false,
"raw_telegram_payload_display_allowed": false,
"private_reasoning_display_allowed": false,
"secret_value_display_allowed": false,
"work_window_transcript_display_allowed": false,
"allowed_display_fields": [
"packet_id",
"display_name",
"risk_tier",
"owner_agent",
"executor_agent",
"executor_type",
"handoff_status",
"controlled_route_id",
"check_mode_ref",
"verifier_ref",
"rollback_ref",
"telegram_evidence_ref",
"km_writeback_ref",
"playbook_trust_ref",
"rollups"
],
"blocked_display_fields": [
"raw tool output",
"raw runtime payload",
"raw Telegram payload",
"private reasoning",
"secret value",
"authorization header",
"work window transcript"
]
},
"rollups": {
"source_readback_count": 8,
"handoff_packet_count": 7,
"ready_for_controlled_executor_count": 5,
"critical_break_glass_count": 2,
"high_risk_packet_count": 5,
"critical_packet_count": 2,
"ansible_check_mode_packet_count": 3,
"mcp_tool_route_count": 7,
"post_action_verifier_binding_count": 5,
"telegram_evidence_binding_count": 5,
"km_writeback_binding_count": 5,
"playbook_trust_writeback_binding_count": 5,
"owner_response_required_count": 2,
"blocked_by_critical_boundary_count": 2,
"missing_check_mode_count": 0,
"missing_rollback_count": 0,
"missing_verifier_count": 0,
"missing_telegram_evidence_count": 0,
"missing_learning_writeback_count": 0,
"executor_route_count": 5,
"verifier_binding_count": 5,
"learning_writeback_contract_count": 3,
"controlled_executor_dispatch_count": 0,
"live_apply_count": 0,
"gateway_queue_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"km_write_count": 0,
"playbook_trust_write_count": 0,
"production_write_count": 0,
"secret_read_count": 0,
"paid_api_call_count": 0,
"host_write_count": 0,
"kubectl_action_count": 0,
"destructive_operation_count": 0
},
"next_actions": [
{
"task_id": "P2-416",
"priority": "P0",
"summary": "建立 controlled executor dispatch worker dry-run，從 P2-415 handoff packet 產生 executor run preview、idempotency key、failure lane 與 verifier queue。",
"gate": "dispatch worker 必須只接受 ready_for_controlled_executor，critical_break_glass_only 仍拒收。"
},
{
"task_id": "P2-417",
"priority": "P0",
"summary": "把 executor receipt 寫回 AwoooP status-chain、日 / 週 / 月報與 Telegram redacted evidence，讓使用者看到每個 Agent 的實際處理量。",
"gate": "receipt 必須有 verifier result、rollback outcome、KM / PlayBook trust writeback ref。"
}
]
}

201
docs/schemas/ai_agent_controlled_executor_handoff_v1.schema.json Normal file
View File

@@ -0,0 +1,201 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "urn:awoooi:ai-agent-controlled-executor-handoff-v1",
"title": "AWOOOI AI Agent controlled executor handoff v1",
"description": "P2-415 承接 P2-409 controlled apply queue，建立 high 風險候選交給受控 executor 前的 handoff readback：allowlist、Ansible check-mode、rollback、post-action verifier、Telegram evidence、KM / PlayBook trust 回寫都必須可讀。此 schema 不授權 live apply、Telegram send、Bot API、secret read、paid API、host write、kubectl 或不可逆操作。",
"type": "object",
"required": [
"schema_version",
"generated_at",
"program_status",
"source_refs",
"source_readbacks",
"handoff_truth",
"executor_handoff_packets",
"executor_routes",
"verifier_bindings",
"learning_writeback_contracts",
"activation_boundaries",
"display_redaction_contract",
"rollups",
"next_actions"
],
"properties": {
"schema_version": { "type": "string", "const": "ai_agent_controlled_executor_handoff_v1" },
"generated_at": { "type": "string", "minLength": 1 },
"program_status": {
"type": "object",
"required": [
"overall_completion_percent",
"current_priority",
"current_task_id",
"next_task_id",
"read_only_mode",
"runtime_authority",
"status_note"
],
"properties": {
"overall_completion_percent": { "type": "integer", "const": 100 },
"current_priority": { "type": "string", "const": "P0" },
"current_task_id": { "type": "string", "const": "P2-415" },
"next_task_id": { "type": "string", "const": "P2-416" },
"read_only_mode": { "type": "boolean", "const": true },
"runtime_authority": { "type": "string", "const": "controlled_executor_handoff_readback_no_live_apply" },
"status_note": { "type": "string", "minLength": 1 }
},
"additionalProperties": false
},
"source_refs": { "type": "array", "minItems": 1, "items": { "type": "string", "minLength": 1 } },
"source_readbacks": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/source_readback" } },
"handoff_truth": { "type": "object", "additionalProperties": { "type": ["boolean", "integer", "string"] } },
"executor_handoff_packets": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/executor_handoff_packet" } },
"executor_routes": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/executor_route" } },
"verifier_bindings": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/verifier_binding" } },
"learning_writeback_contracts": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/learning_writeback_contract" } },
"activation_boundaries": { "type": "object", "additionalProperties": { "type": "boolean" } },
"display_redaction_contract": { "type": "object", "additionalProperties": { "type": ["boolean", "array", "string"] } },
"rollups": { "type": "object", "additionalProperties": { "type": "integer" } },
"next_actions": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/next_action" } }
},
"$defs": {
"source_readback": {
"type": "object",
"required": ["readback_id", "source_schema_version", "source_ref", "endpoint", "owner_agent", "status", "key_readback", "next_action"],
"properties": {
"readback_id": { "type": "string", "minLength": 1 },
"source_schema_version": { "type": "string", "minLength": 1 },
"source_ref": { "type": "string", "minLength": 1 },
"endpoint": { "type": "string", "minLength": 1 },
"owner_agent": { "type": "string", "enum": ["openclaw", "hermes", "nemotron", "sre", "security", "devops"] },
"status": { "type": "string", "minLength": 1 },
"key_readback": { "type": "string", "minLength": 1 },
"next_action": { "type": "string", "minLength": 1 }
},
"additionalProperties": false
},
"executor_handoff_packet": {
"type": "object",
"required": [
"packet_id",
"source_queue_item_id",
"display_name",
"risk_tier",
"owner_agent",
"executor_agent",
"executor_type",
"handoff_status",
"controlled_route_id",
"playbook_ref",
"mcp_tool_ref",
"check_mode_ref",
"verifier_ref",
"rollback_ref",
"telegram_evidence_ref",
"km_writeback_ref",
"playbook_trust_ref",
"allowlist_match",
"check_mode_passed",
"rollback_plan_ready",
"post_action_verifier_ready",
"telegram_evidence_ready",
"km_writeback_ready",
"playbook_trust_writeback_ready",
"owner_response_required",
"break_glass_required",
"controlled_executor_handoff_allowed",
"live_apply_performed",
"side_effect_count",
"blocked_runtime_actions",
"next_gate"
],
"properties": {
"packet_id": { "type": "string", "minLength": 1 },
"source_queue_item_id": { "type": "string", "minLength": 1 },
"display_name": { "type": "string", "minLength": 1 },
"risk_tier": { "type": "string", "enum": ["high", "critical"] },
"owner_agent": { "type": "string", "enum": ["openclaw", "hermes", "nemotron", "sre", "security", "devops"] },
"executor_agent": { "type": "string", "enum": ["openclaw", "hermes", "nemotron", "sre", "security", "devops"] },
"executor_type": { "type": "string", "enum": ["ansible_playbook", "mcp_tool_route", "telegram_gateway_queue", "km_playbook_writer", "readback_verifier", "break_glass_only"] },
"handoff_status": { "type": "string", "enum": ["ready_for_controlled_executor", "critical_break_glass_only", "blocked_missing_check_mode", "blocked_missing_verifier", "blocked_missing_learning_writeback"] },
"controlled_route_id": { "type": "string", "minLength": 1 },
"playbook_ref": { "type": "string", "minLength": 1 },
"mcp_tool_ref": { "type": "string", "minLength": 1 },
"check_mode_ref": { "type": "string", "minLength": 1 },
"verifier_ref": { "type": "string", "minLength": 1 },
"rollback_ref": { "type": "string", "minLength": 1 },
"telegram_evidence_ref": { "type": "string", "minLength": 1 },
"km_writeback_ref": { "type": "string", "minLength": 1 },
"playbook_trust_ref": { "type": "string", "minLength": 1 },
"allowlist_match": { "type": "boolean" },
"check_mode_passed": { "type": "boolean" },
"rollback_plan_ready": { "type": "boolean" },
"post_action_verifier_ready": { "type": "boolean" },
"telegram_evidence_ready": { "type": "boolean" },
"km_writeback_ready": { "type": "boolean" },
"playbook_trust_writeback_ready": { "type": "boolean" },
"owner_response_required": { "type": "boolean" },
"break_glass_required": { "type": "boolean" },
"controlled_executor_handoff_allowed": { "type": "boolean" },
"live_apply_performed": { "type": "boolean", "const": false },
"side_effect_count": { "type": "integer", "const": 0 },
"blocked_runtime_actions": { "type": "array", "minItems": 1, "items": { "type": "string" } },
"next_gate": { "type": "string", "minLength": 1 }
},
"additionalProperties": false
},
"executor_route": {
"type": "object",
"required": ["route_id", "display_name", "executor_agent", "route_status", "required_inputs", "blocked_actions", "live_apply_allowed_by_this_readback"],
"properties": {
"route_id": { "type": "string", "minLength": 1 },
"display_name": { "type": "string", "minLength": 1 },
"executor_agent": { "type": "string", "enum": ["openclaw", "hermes", "nemotron", "sre", "security", "devops"] },
"route_status": { "type": "string", "enum": ["ready_for_handoff", "blocked_by_policy"] },
"required_inputs": { "type": "array", "minItems": 1, "items": { "type": "string" } },
"blocked_actions": { "type": "array", "minItems": 1, "items": { "type": "string" } },
"live_apply_allowed_by_this_readback": { "type": "boolean", "const": false }
},
"additionalProperties": false
},
"verifier_binding": {
"type": "object",
"required": ["binding_id", "display_name", "owner_agent", "required_before_dispatch", "ready_count", "blocked_count", "failure_if_missing"],
"properties": {
"binding_id": { "type": "string", "minLength": 1 },
"display_name": { "type": "string", "minLength": 1 },
"owner_agent": { "type": "string", "enum": ["openclaw", "hermes", "nemotron", "sre", "security", "devops"] },
"required_before_dispatch": { "type": "boolean", "const": true },
"ready_count": { "type": "integer", "minimum": 0 },
"blocked_count": { "type": "integer", "minimum": 0 },
"failure_if_missing": { "type": "string", "minLength": 1 }
},
"additionalProperties": false
},
"learning_writeback_contract": {
"type": "object",
"required": ["contract_id", "display_name", "owner_agent", "target_store", "writeback_status", "required_fields", "runtime_write_performed"],
"properties": {
"contract_id": { "type": "string", "minLength": 1 },
"display_name": { "type": "string", "minLength": 1 },
"owner_agent": { "type": "string", "enum": ["openclaw", "hermes", "nemotron", "sre", "security", "devops"] },
"target_store": { "type": "string", "minLength": 1 },
"writeback_status": { "type": "string", "enum": ["ready_for_executor_receipt", "blocked_by_policy"] },
"required_fields": { "type": "array", "minItems": 1, "items": { "type": "string" } },
"runtime_write_performed": { "type": "boolean", "const": false }
},
"additionalProperties": false
},
"next_action": {
"type": "object",
"required": ["task_id", "priority", "summary", "gate"],
"properties": {
"task_id": { "type": "string", "minLength": 1 },
"priority": { "type": "string", "minLength": 1 },
"summary": { "type": "string", "minLength": 1 },
"gate": { "type": "string", "minLength": 1 }
},
"additionalProperties": false
}
},
"additionalProperties": false
}