| 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 โ including Korean phrases like ์๋น์ค ํํฉ, ์๋ฌ ๋ฆฌํฌํธ, ์ฅ์ ํํฉ, latency ๋ถ์, API ์ฑ๋ฅ, or even casual requests like 'thomas ์ด์ ์๋ฌ ์ข ๋ด์ค' 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 โ maps to Datadog
service tag (e.g., Thomas, Alice)
- Time range โ start/end in KST (end defaults to now)
- Environment โ prod/dev/stage (defaults to prod)
Execution
1. Parse parameters
Convert timestamps to milliseconds using Python with KST (UTC+9) โ 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 โ 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 โ 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 โ without it, infra noise buries real application errors.
| Category | What it is | Action |
|---|
| A โ Infra noise | JVM/agent artifacts misclassified as errors | Report separately, don't count as app errors |
| B โ Deploy-related | Transient errors during rolling updates | Correlate with deploy events, report separately |
| C โ Application errors | Everything else | Core of the report โ 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 โ special characters break the UI. See references/datadog-queries.md for the URL format.
