tune-ai-powered-devops
/tune-ai-powered-devops
Dostraja narzędzia AIOps: okna korelacji alertów, wagi ryzyka, pewność remediacji i koszt przetwarzania.
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 26.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Dostraja narzędzia AIOps: okna korelacji alertów, wagi ryzyka, pewność remediacji i koszt przetwarzania.
Oryginalny opis (EN)
Tune AI-Powered DevOps — optimize alert correlation windows, risk scoring weights, remediation confidence, incident prompt quality, cost. Use when: tune, optimize AIOps.
Jak działa
- When to Use
- Tuning Dimensions
- Production Readiness Checklist
- Output: Tuning Report
- Tuning Playbook
Podgląd SKILL.md
--- name: tune-ai-powered-devops description: "Tune AI-Powered DevOps — optimize alert correlation windows, risk scoring weights, remediation confidence, incident prompt quality, cost. Use when: tune, optimize AIOps." --- # Tune AI-Powered DevOps ## When to Use - Optimize alert correlation time windows and grouping - Calibrate risk scoring weights from deploy history - Tune auto-remediation confidence thresholds - Improve LLM incident analysis prompt quality - Reduce false positives in risk scoring ## Tuning Dimensions ### Dimension 1: Alert Correlation Tuning | Parameter | Default | Range | Impact | |-----------|---------|-------|--------| | Time window | 5 min | 2-15 min | Wider = more grouped, narrower = more incidents | | Service grouping | Same service | Service chain | Chain = catches cascading failures | | Max alerts per incident | 100 | 20-500 | Too low = splits real incidents | | Dedup window | 1 min | 30s-5 min | Prevents repeat alerts counting as separate | **Tuning rule**: Start at 5 min. If incidents split → widen. If unrelated grouped → narrow. ### Dimension 2: Risk Scoring Weight Calibration | Factor | Default | After 30 Deploys | After 100 Deploys | |--------|---------|-----------------|-------------------| | Change size | 0.25 | Adjust from data | Adjust from data | | Blast radius | 0.30 | Adjust from data | Adjust from data | | Time of day | 0.10 | Adjust from data | Adjust from data | | Author exp | 0.10 | Adjust from data | Adjust from data | | Test coverage | 0.15 | Adjust from data | Adjust from data | | Dependencies | 0.10 | Adjust from data | Adjust from data | **Method**: Compare predicted risk vs actual outcome. Increase weight for factors that predict failures well. ### Dimension 3: Remediation Confidence | Action Type | Default Threshold | Conservative | Aggressive | |------------|------------------|-------------|-----------| | Restart pod | 0.95 | 0.99 | 0.90 | | Scale up | 0.90 | 0.95 | 0.85 | | Rotate certificate | 0.99 | 0.99 | 0.95 | | Archive logs | 0.85 | 0.90 | 0.80 | | Rollback deploy | 0.80 | 0.95 | 0.70 | | Database fix | N/A (always human) | N/A | N/A | ### Dimension 4: Incident Analysis Prompt | Optimization | Before | After | Impact | |-------------|--------|-------|--------| | Add logs to context | Generic analysis | "Based on these error logs..." | +30% accuracy | | Add deployment timeline | "Something failed" | "Deploy at 14:00, errors at 14:05" | +25% root cause | | Add dependency map | Isolated analysis | "Service A depends on Service B" | +20% cascade detection | | Add past remediation | Suggests new fix | "Last time this was fixed by..." | +40% action quality | ### Dimension 5: Cost Per Incident | Component | Cost | Optimization | |-----------|------|-------------| | LLM analysis (gpt-4o) | ~$0.05/incident | Use gpt-4o-mini for triage, gpt-4o for full analysis | | Alert processing | ~$0.001/alert | Batch process, correlate before LLM | | Risk scoring | ~$0.01/deploy | Cache for same commit | | Auto-remediation | ~$0.001/action | Only Azure API calls | **Monthly estimate** (50 incidents/mo, 200 deploys/mo): - Incident analysis: ~$2.50/mo - Risk scoring: ~$2.00/mo - Auto-remediation: negligible - **Total: ~$5/mo** for AI cost (infrastructure cost separate) ## Production Readiness Checklist - [ ] Alert correlation grouping correctly (≥ 90%) - [ ] Risk scoring calibrated (±1 of actual outcome) - [ ] Auto-remediation tested for all safe actions - [ ] Human approval gate for risky remediations - [ ] MTTR improved ≥ 40% vs manual - [ ] Post-incident reports actionable - [ ] Notification channels connected - [ ] Incident history feeding back into analysis - [ ] Blast radius check preventing multi-service impact ## Output: Tuning Report After tuning, compare: - Alert correlation improvement (noise reduction %) - Risk score calibration delta - Auto-remediation success rate - MTTR reduction - False positive rate change ## Tuning Playbook 1. **Baseline**: Replay 20 past incidents, record LLM accuracy 2. **Correlation**: Adjust time window based on actual incident patterns 3. **Risk weights**: Compare 30 deploys: predicted vs actual outcome 4. **Remediation**: Start conservative (0.95), lower for proven safe actions 5. **Prompt**: Add logs + deploy timeline + dependency map to analysis 6. **MTTR**: Measure time-to-resolve with vs without AI assistance 7. **Cost**: Confirm AI cost < $5/mo (negligible vs ops time saved) 8. **Re-test**: Same 20 incidents, compare before/after
Pliki w paczce (3)
- SKILL.lean.md 4 KB
- SKILL.md 4 KB
- agents/openai.yaml 1 KB
Wersje
W bazie nie ma innych skilli o nazwie tune-ai-powered-devops.
Migawka bazy z 13.11.2026 · wersja paczki 37a9f5cf343e (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100deploy-ai-powered-devops
Wdraża DevOps wspierany przez AI: pipeline incydentów, ocenę ryzyka wdrożeń, runbooki auto-remediacji, korelację alertów i analizę wpływu zmian. - 100evaluate-ai-powered-devops
Ocenia DevOps wspierany przez AI: trafność reakcji na incydenty, kalibrację oceny ryzyka, bezpieczeństwo automatycznej naprawy, jakość korelacji alertów i skrócenie MTTR. - 100openocta
Prowadzi OpenOcta, agenta AIOps do analizy alertów, zapytań o dane i napraw, a także konfiguracji modeli i instalacji. - 90llm-aiops
Zbiera przegląd badań o podejściach opartych na LLM do operacji IT: zarządzanie incydentami, analiza przyczyn źródłowych, parsowanie logów, wykrywanie anomalii, IaC i diagnoza wydajności. - 100pr-quality-devops
Ocenia lokalny diff pod kątem ryzyka CI/CD, pipeline'u budowania i infrastruktury, z werdyktem PASS, WARN lub CRITICAL_FAIL. - 100pr-quality-devops
Ocenia lokalny diff pod kątem ryzyka dla CI/CD, pipeline'u budowania i infrastruktury, zwracając werdykt PASS, WARN lub CRITICAL_FAIL.