OG T
|
95f63d64d7
|
fix(auto_approve): min_trust_score 0 解除自動修復封鎖
CD Pipeline / build-and-deploy (push) Has been cancelled
根本原因: trust_score 是 in-memory dict，Pod 重啟即歸零
永遠 < min_trust_score=1 → 所有告警走審批，從未自動執行
修復: min_trust_score=0，medium risk + confidence>=0.65 直接自動執行
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-10 16:06:40 +08:00 |
|
OG T
|
4c622813af
|
fix(auto-repair): 實際可用的自動修復門檻 (Phase 22 P1)
E2E Health Check / e2e-health (push) Has been cancelled
CD Pipeline / build-and-deploy (push) Has been cancelled
問題: 四道鎖全卡死導致自動修復永遠不觸發
1. configmap: Gemini 排第一 (100ms vs NVIDIA 60s timeout)
2. auto_approve: confidence 0.90→0.65, trust 5→1, playbook 3→1
3. auto_approve: 開放 medium 風險, require_playbook=False
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-01 16:02:16 +08:00 |
|
OG T
|
938df7f291
|
fix(api): 全面清除假信心分數 - 遵循 feedback_confidence_truthfulness.md
🔴 違規修正: 規則匹配/Expert System 不是 AI 分析，confidence 必須 = 0.0
修正檔案:
- agents/action_planner.py: 0.9 → 0.0
- agents/blast_radius.py: 0.85/0.5/0.9 → 0.0
- agents/security.py: 計算公式 → 0.0
- signoz_webhook.py: 0.7 → 0.0
- auto_approve.py: default 0.5 → 0.0
- ci_auto_repair.py: 整個計算函數 → return 0.0
- error_analyzer_service.py: default 0.5 → 0.0
- intent_classifier.py: 計算公式 → 0.0
- openclaw.py: default 0.5 → 0.0
- resource_resolver.py: 0.8 → 0.0
- k8s_naming.py: 0.9/0.7 → 0.0
只有 LLM 真實分析返回的 confidence 才能 > 0
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
|
2026-03-29 16:00:46 +08:00 |
|
OG T
|
138ef0c2db
|
fix(api): 修復 7 個 Lint 錯誤 (unused imports + zip strict + dict comprehension)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
|
2026-03-27 14:42:47 +08:00 |
|
OG T
|
ce7f8a1b23
|
feat(api): ADR-030 Phase 4 自動執行機制
實作低風險操作自動執行策略：
1. auto_approve.py - 自動執行策略服務
- AutoApprovePolicy: 評估是否可自動執行
- 條件: LOW 風險 + 信任分數 >= 5 + Playbook 成功率 >= 95%
- CRITICAL 永遠不自動執行
- 完整審計追蹤
2. trust_engine.py - 新增 singleton
- get_trust_manager(): 取得全域 TrustScoreManager
3. decision_manager.py - 整合自動執行 (Tier 3 紅區)
- Step 5 加入 AutoApprovePolicy 判斷
- 條件滿足時跳過 Telegram，直接執行
- _auto_execute(): 自動執行邏輯
- 失敗時 fallback 到人工審核
流程:
Incident → 分析 → AutoApprovePolicy 評估
├─ 可自動執行 → 直接執行 → 完成
└─ 需人工審核 → Telegram 通知 → 等待批准
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
|
2026-03-26 22:13:10 +08:00 |
|