菜单
Guides agents through production data incidents. Use when a pipeline fails, publishes bad data, misses an SLA, partially loads, corrupts state, or requires rollback, replay, or stakeholder communication.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Forces replay-safe rollout plans, reconciliation gates, and rollback paths before executing any data backfill or historical reprocessing. Use when running /backfill, rerunning pipelines, repairing publish windows, or any work that risks double-counting or downstream corruption.
Helps agents classify data engineering work, choose the right preset and skill bundle, and pick the safest next command. Use when starting a session, triaging an ambiguous request, or deciding how to proceed.
Guides agents through resiliency testing for data platforms. Use when designing or running failure drills, recovery validation, failover tests, replay-safety checks, dependency outage exercises, or fault injection for pipelines and publishes.
Enforces production Kafka guardrails including non-breaking schema evolution, dead-letter queues for poison messages, and acks=all producer durability. Use when designing or changing Kafka topics, producers, consumers, schema registry policies, or streaming recovery paths.
Guides agents to wire Model Context Protocol servers for live data platform observability including Spark execution plans, OOM diagnosis, Kafka consumer lag, and orchestration run state. Use when agents need cluster metrics, streaming lag, or job diagnostics instead of blind code changes.
Designs scheduling, reruns, and backfills safely for data systems. Use when changing orchestration, retries, dependency timing, historical reprocessing, or publish sequencing.
| name | incident-triage-and-pipeline-recovery |
| description | Guides agents through production data incidents. Use when a pipeline fails, publishes bad data, misses an SLA, partially loads, corrupts state, or requires rollback, replay, or stakeholder communication. |
Use this skill when production behavior is already wrong and the team needs controlled recovery. It helps agents contain blast radius, diagnose quickly, restore trust, and avoid making the incident worse.
Do not jump to fixes before stabilizing the system and understanding impact.
Contain the incident. Decide whether to:
Pull live signals before mutation when possible.
Load mcp-data-observability-integration to inspect lag, run state, or Spark stage metrics before reruns or replays.
Classify impact. Identify:
Determine the safest recovery path. Options include:
When replay or backfill is chosen, load safe-backfill-and-replay-orchestration and draft templates/backfill-plan.yaml before execution.
Validate recovery before reopening publish paths.
Record the incident and add a guardrail. A fixed incident with no prevention work is unfinished.
Turn high-value incidents into repeatable resilience tests.
Load skills/data-resiliency-testing-and-failure-injection/SKILL.md or references/data-resiliency-testing-patterns.md when the same failure mode must be prevented from surprising the team again.
| Rationalization | Reality |
|---|---|
| "Let us rerun everything now." | Blind reruns can duplicate data or destroy evidence. |
| "We can clean up downstream later." | Downstream trust loss is often harder to recover than the pipeline itself. |
| "The root cause can wait." | Without a guardrail, the same incident often returns quickly. |
