Name: Production Incident Commander
Brand: Tetrees AI
Availability: InStock

Tetrees 에이전트 가이드

이 에이전트가 하는 일

During an outage, responders jump between alerts, logs, metrics, traces, topology, and recent changes; weak timelines and untested hunches create noisy investigations and risky remediation.

권장 사용자

On-call engineers and incident commanders
SRE and platform teams
Teams building runbooks and synthetic incidents
Engineering managers reviewing postmortems

사용하는 입력

Alert, symptom, severity, and incident time window
Metrics, logs, traces, topology, deploys, and events
Service objectives, ownership, and runbooks
Approved-action boundaries and rollback options

제공하는 결과

Incident timeline and blast-radius map
Ranked hypothesis table with evidence
Reversible mitigation and rollback plan
Root-cause and contributing-factor report
Postmortem action and monitoring checklist

작업 진행 방식

1Establish incident scope, customer impact, severity, time window, and immediate safety constraints.
2Build a normalized timeline from independent observability and change sources.
3Rank hypotheses by explanatory coverage, temporal fit, topology, and contradicting evidence.
4Separate symptom relief from root-cause repair and choose the safest reversible mitigation.
5Require owner confirmation before any write, production, or customer-impacting action.
6Produce an evidence-linked incident report, residual-risk watch plan, and durable learning candidates.

사용 예시

아래 요청을 시작점으로 삼고 세부 내용을 상황에 맞게 바꾸세요.

Diagnose a database connection storm

API latency and error rate rose after a deployment while database CPU and connection metrics disagree.

예시 요청

Use these alert, CloudWatch metrics, deploy events, and application logs to rank causes. Distinguish DB saturation, connection leakage, and retry amplification, then propose the safest mitigation.

예상 결과: A normalized timeline, correlated evidence, ranked hypotheses, disconfirming checks, a reversible mitigation, rollback criteria, and no unapproved production action.

Review whether a noisy alert is causal

Several infrastructure alerts fire together and the team risks treating the loudest one as the root cause.

예시 요청

Analyze this incident bundle. Identify which alerts are causal, consequential, coincidental, or unsupported, and give the next three evidence-gathering steps.

예상 결과: A topology- and time-aware correlation report, evidence gaps, confidence, safe next checks, and a bounded incident status update.

성공적인 실행 기준

✓ Timeline and topology support the root-cause claim
✓ Counterevidence is recorded
✓ Mitigation is reversible and has rollback criteria
✓ No production write occurs without explicit approval
✓ Postmortem actions are specific and testable

제약 및 한계

• Hosted mode is read-only and cannot mutate infrastructure
• Incomplete telemetry can prevent a confident diagnosis
• Correlation is not proof of causation
• Human incident command remains responsible for production decisions

기능

incident triageevidence correlationroot cause analysismitigation planningpostmortem

Agent Studio에서 OpenAI, Claude 또는 Z.AI를 선택하고 Tetrees 포인트나 BYOK로 실행하세요.

실행 도구

호스팅 실행은 서명된 도구만 사용합니다. 로컬 MCP에서는 별도로 설정하고 승인한 도구를 추가할 수 있습니다.

recall pack memorypropose pack growthlist run attachmentsread run attachmentweb search

Agent AVCP

Agent AVCP 리포트

테스트 버전 v1.0.0

테스트한 TAIP 아티팩트는 이 리포트와 플랫폼 서명에 결합됩니다. 점수는 테스트 근거이며 모든 향후 모델 응답을 보장하지 않습니다.

종합 점수

9.5

10점 만점

보안9.5

기능10.0

완성도10.0

필수 게이트

provenance✓ 통과

costCeilings✓ 통과

growthRollback✓ 통과

skillExecution✓ 통과

memoryIsolation✓ 통과

이 팩을 실행하거나 추가하려면 로그인하세요.

리뷰

아직 리뷰가 없습니다.

Production Incident Commander