logdet-triagelisted
Install: claude install-skill vivekbhange-hub/log-detective-skill
# logdet-triage: interrogate the log, do not read it
A production log is millions of lines and a handful of distinct events. This skill runs
`log_cluster.py` over the file, so the conversation only ever sees the ranked signatures, their
timing and one sample each, then turns that into a first hypothesis you can test.
## Inputs
- A log path (plain text, JSON lines, logfmt, syslog, access log; `.gz` works, `-` reads stdin).
Several paths are clustered together.
- Optional: the time window of interest, the service name, the level to start from.
- Commands below say `python3`; on Windows use `python` (or `py -3`).
## Outputs
- `logs-triage/clusters.md` (or whatever path the user prefers): the report from the script.
- In chat: the top 3 to 5 signatures with counts and first-seen times, a rate observation, and
one hypothesis in the fixed shape from Step 6.
## Steps
1. **Confirm authorization.** Ask once: are these logs from a system the user owns, runs or has
permission to analyse? Logs often contain personal data; if they are someone else's, stop.
2. **Do not open the file.** Do not `cat`, `head` or Read the log. If you need its size, use
`ls -l` (Windows: `dir`) rather than counting lines.
3. **Run the clustering**, errors first:
```
python3 "${CLAUDE_SKILL_DIR}/log_cluster.py" app.log --level error --top 15 --redact \
-o logs-triage/clusters.md
```
- `--redact` masks emails, tokens, JWTs, bearer headers, IPs and card-like numbers in what you