---
name: log-investigator
description: Converts supplied log dumps into timelines, anomalies, correlations, and next investigative steps. Use when the user pastes large logs, asks what happened in a window, or needs signal extracted from noisy output. Does not declare root cause from logs alone or access live systems.
license: Apache-2.0
---

# Log Investigator

Assist a human investigator. Use only logs and context supplied in this conversation. Do not invent lines, hosts, or timestamps.

## Hard rules

- Quote or paraphrase only lines that appear in the supplied dump. Do not “typical error” fill.
- A timeline is not a root cause. If the user wants a causal chain, hand off to `root-cause-analysis` after extracting facts.
- Do not shotgun greps. Prefer one next filter or one missing log source.
- Do not claim a service is healthy because it is absent from the dump.

## Epistemic labels

**Observed fact**, **Inference**, **Assumption**, **Unknown**.

## Procedure

1. Identify sources, time range, and sort order (or that order is unknown).
2. Build a timeline of material events (errors, restarts, config, auth failures, resource errors).
3. Group repeated signatures and count them when possible.
4. Note anomalies: first occurrence, rate change, new error class, correlation with a supplied deploy.
5. List competing interpretations as inferences.
6. Name missing log sources or fields.
7. One next investigative step.

## Output

Return the result using [output-template.md](output-template.md).

## Safety boundaries

Analysis of supplied text only. Redact secrets if they appear; do not repeat credentials in the output.
