Audit how you use AI in Claude Code. The scan runs on your machine. The report you upload is mostly counts and scores — but it also carries a few short quoted excerpts from your own sessions. What gets sent.
One Python file. No dependencies beyond the standard library.
Reads ~/.claude/projects/*.jsonl. Your session files stay put — the report holds counts, scores, and up to 5 short excerpts per module.
Drop or paste the JSON. We score server-side and render your dashboard.
Then run from where the browser saved it (usually
~/Downloads):
cd ~/Downloads && python3 supermark-observability-audit-0.11.1.py --comprehension-strip --json > supermark-obs-report.json
--json writes clean JSON to stdout and nothing to stderr — safe to redirect, pipe, or paste raw.
--comprehension-strip replaces paths, slugs and first names inside the quoted excerpts with placeholders before they are written to the file.
Open the file and read it before you upload it — it is plain JSON, and it is yours.
or click to browse
Last updated 2026-07-30.
Mark Turrell, operating SuperMark, is the data controller. Questions or requests: projectneuland@gmail.com.
The script reads your local Claude Code session files. Most of what
it uploads is counts and scores. It is not only aggregates —
the Comprehension module also includes up to 5 short quoted excerpts
per module (up to 240 characters of assistant text, 200 of your own)
so it can show you the moments it scored. Running it with
--comprehension-strip, as the command on this page
does, replaces paths, slugs and first names in those excerpts with placeholders
first. Without that flag the excerpts are sent verbatim, and the report records
which of the two it was. In --scan-all mode the
report also lists your project directory names.
The file is plain JSON on your own machine. Open it and read it before you upload it. Nothing is transmitted until you choose to upload.
Your audit is processed to produce and return the dashboard you asked for — that is the service you requested, and the basis is performance of that request (GDPR Art 6(1)(b)). Aggregated, non-identifying results are also used to improve the audit method itself (Art 6(1)(f), legitimate interest). It is not sold, and it is not used to train models.
None are collected. This page previously offered an optional contact field; it was withdrawn on 2026-07-30 and will not return until there is a proper consent step, a record of that consent, and a retention rule to go with it.
Your audit is kept until you delete it. There is currently no automatic expiry — stating that plainly rather than implying a limit that does not exist. An event-anchored retention rule is being implemented.
Your retrieval token is the only thing that identifies your audit — no account, no email. Load your dashboard with the token and use Delete my audit, which erases the stored audit, any personal-review request and any shared summary. If you lose the token, that record cannot be found or deleted, by you or by us. Keep it, or delete the audit when you are done.