Cluster a batch of error log lines into root-cause groups with counts and likely causes.

Log Error Clusterer

Turns a noisy pile of error log lines into a short, ranked list of root-cause groups — so you can see what is failing and how often instead of scrolling thousands of near-duplicate lines.

What it does

The InTouch AI assistant loads this skill as a prompt. It:

Splits the batch into individual entries (multi-line stack traces stay attached to their parent line).
Normalizes each entry into a stable signature by masking the variable parts — timestamps, IDs, UUIDs, paths, IPs, ports, line numbers, hex addresses, quoted literals — leaving the exception class, error code, and message skeleton.
Clusters entries with matching signatures (same exception class + top stack frame counts as one cluster).
Counts each cluster, ranks by frequency (severity breaks ties), and infers a likely root cause plus a suggested next step for each.

What to feed it

A batch of raw error/warning log lines as plain text — for example the tail of an application log, a grep ERROR dump, a CI failure log, or aggregated alert output. One entry per line; stack traces may follow their line. No particular format is required; the skill tolerates mixed timestamp styles and unstructured messages.

What it returns

A Markdown Error Log Cluster Report:

Total entries parsed and number of clusters.
One section per cluster (ordered by count, then severity): short name, masked signature, likely cause, suggested next step, and a representative example line.
An Unclassified count for anything that couldn't be grouped.
A Summary naming the top cluster (with percentage) and the highest-severity cluster.

It will not invent error codes or stack frames that aren't in the input; uncertain clusters are flagged "needs investigation" rather than guessed.

Example use

Paste a few hundred lines of an application error log and ask the assistant to run log-error-clusterer. Instead of reading every line, you get back something like "78% of errors are a single DB connection-pool-exhausted cluster; one FATAL cluster is a null deref in OrderService" with the next step for each.