| name | datadog-error-report |
| description | Generate a comprehensive error status report for a service instance using Datadog MCP. Use this skill whenever the user asks about error reports, error status, service health, latency analysis, API performance, incident checks — including Korean phrases like 서비스 현황, 에러 리포트, 장애 현황, latency 분석, API 성능, or even casual requests like 'thomas 어제 에러 좀 봐줘' or 'check alice errors last week'. If the user mentions a service name and wants to know what went wrong, this is the right skill. |
Datadog Error Report
Collect error logs, incidents, monitors, events, and API performance metrics from Datadog MCP, classify noise from real errors, and produce a structured report.
Input
The user provides:
- Service name — maps to Datadog
service tag (e.g., Thomas, Alice)
- Time range — start/end in KST (end defaults to now)
- Environment — prod/dev/stage (defaults to prod)
Execution
1. Parse parameters
Convert timestamps to milliseconds using Python with KST (UTC+9) — seconds will silently produce wrong results. See references/datadog-queries.md for the conversion snippet.
2. Load Datadog MCP tools
Run 6 ToolSearch calls in parallel before any data queries — the tools aren't available until loaded. See references/datadog-queries.md for the exact tool names.
3. Collect data in parallel (6 queries)
Launch all simultaneously — running them sequentially takes 6x longer for no benefit:
| # | Data | Purpose |
|---|
| 1 | Error logs (pattern mode) | See the big picture of all errors |
| 2 | Incidents | Active/resolved incidents for the service |
| 3 | Monitors | Alert/warn/OK status |
| 4 | Events | Deploys, scaling, incidents |
| 5 | p99 latency by endpoint | Find slow APIs |
| 6 | TPS / request volume | Understand traffic patterns |
See references/datadog-queries.md for exact query templates.
4. Classify error logs (A/B/C)
This is the most important step — without it, infra noise buries real application errors.
| Category | What it is | Action |
|---|
| A — Infra noise | JVM/agent artifacts misclassified as errors | Report separately, don't count as app errors |
| B — Deploy-related | Transient errors during rolling updates | Correlate with deploy events, report separately |
| C — Application errors | Everything else | Core of the report — analyze in detail |
See references/noise-patterns.md for the full pattern list and exclusion filters.
5. Fetch raw logs for Category C
Re-query with noise exclusion filters to get actionable detail on real application errors.
6. Generate report
Use the template in references/report-template.md. The report covers:
- Error summary with A/B/C classification
- Category C detail with frequency, time distribution, and causal chain analysis
- Incidents, monitors, events
- API performance (TPS, p99, slow APIs)
- Prioritized recommendations
7. Deep links
When the user wants to drill into specific errors, construct Datadog URLs using facet queries (@error.message:*keyword*), not full-text search — special characters break the UI. See references/datadog-queries.md for the URL format.
