Community

anomaly-detection

Open Agent Skill · data · works with Claude Code, Codex, Cursor, and compatible agents

Investigate metric and log anomalies with structured hypotheses and next checks. Use whenever the user sees a spike, drop, or weird pattern in metrics or logs and asks what is going on, whether it is real, or how to investigate.

View on GitHub

Anomaly Detection (investigation)

Start with “is it real?”, then scope, then hypotheses, then disconfirm. Avoid declaring root cause from one graph.

Workflow

  1. Define the signal, window, and expected baseline.
  2. Check instrumentation changes, deploys, traffic, and batch jobs.
  3. Segment: region, endpoint, tenant, version.
  4. Correlate logs/traces/events (not only CPU).
  5. Rank hypotheses; suggest next 2–3 checks.
  6. If alerting: propose better signal vs noise.

Output format

## Anomaly read: <signal>
**Real?** likely|unclear|artifact
**Scope:** …
**Hypotheses:** 1… 
**Next checks:** …
**Mitigations if user-impacting:** …

Rules

  1. Do not invent timestamps or deploy times.
  2. Distinguish level shift vs spike vs missing data.
  3. Seasonality and weekdays matter.
  4. Prefer actionable next steps over ML buzzwords.
  5. Security anomalies → involve security path.

Edge cases

  • Counter resets / rate math errors.
  • Canary vs full traffic.
  • Silent dashboard timezone bugs.