skip to content

~/security

A secret scanner is a target. We built it like one.

The safest secret is one we never hold. Leaksweep is designed so that a full credential exists only in memory, only for the length of one request, and nowhere else.

data handling

What happens to a secret we find.

masking at source

Masked inside the scanner, before anything else sees it.

The detector that matches a candidate is the same code that masks it. Display, storage, logs, and the reasoning layer all receive the masked form. There is no reveal toggle, no admin view, and no debug flag that changes this.

length rule example
12+ chars first 4 + last 4 sk_l••••••••YhJ9
8 to 11 first 2 + last 2 hu••••••42
under 8 nothing survives ••••••••

Excerpt lines get a second pass: any other long high-entropy token on the line is masked too, in case a detector missed something that still should not be echoed back.

no raw secret storage

masked       AKIA••••••••5RWD
fingerprint  hmac-sha256, keyed,
             truncated
location     config/aws.ts:14

That is the whole record. The fingerprint key lives server side, so the value can not be reversed or brute-forced from our storage.

no source storage

Pasted code, uploaded archives, and fetched repos are processed in memory and dropped when the request ends. Nothing is written to disk or object storage. What persists is a redacted excerpt of a few lines around each finding.

what the model sees

Masked context only.

The reasoning layer gets the detector name, the masked value, entropy, the file path, and a redacted excerpt. It never gets the full secret and never gets the whole file. Model output is parsed against a schema and treated as untrusted input. Your data is not used to train models, ours or a provider's.

never validated

We do not test credentials. Ever.

Leaksweep never sends a found credential to the service that issued it. Validation is indistinguishable from use, and use of a credential we do not own is not defensive work. Risk is judged from context, not from a live call.

platform

How the service is locked down.

encryption

in transit
TLS 1.2+
at rest
AES-256 (provider)
plain http
redirected to https

Calls to a model provider go over TLS as well, carrying masked context only.

authentication

  • >Passwords hashed with scrypt, per-user random salt, compared in constant time.
  • >Sessions are signed JWTs in an HttpOnly, Secure, SameSite cookie.
  • >GitHub and Google sign-in over OAuth. We request identity scopes only, never repo access.
  • >Password reset links expire and are rate limited per caller.

tenant isolation

Every read and every write of a sweep checks that the session user owns it. A valid ID that belongs to someone else returns the same 404 as an ID that does not exist, so sweep IDs can not be probed.

hostile input: repo fetch and archives

SSRF-safe repo fetch.

A repo URL is parsed, the host is checked against an allowlist of two, and the download URL is rebuilt from the parsed owner and repo name. The string you typed is never fetched, so it can not be pointed at an internal address, another host, or another scheme. Private repos are not fetched at all: upload an archive.

Zip-bomb guards.

Each entry is checked against the per-file cap before it is inflated, the running total of extracted bytes is capped, and a tarball that expands past 64 MB is rejected outright. Entries are never written to disk, so path traversal has nowhere to land.

repo hosts
github.com, gitlab.com only
url handling
parsed, then rebuilt from parts
repo download cap
30 MB, stream aborted past it
per-file cap
512 KB, larger files skipped
extracted total cap
12 MB
file count cap
3,000
tar.gz expansion cap
64 MB, rejected past it
binary files
skipped

rate limits and size caps

Auth, form, and sweep routes are rate limited per caller. Sweeps are also limited per account in a rolling ten minute window. Input size is enforced on the server by plan, before any parsing starts.

paste    200 KB free, 1 MB paid
archive  1 MB free, 4 MB paid

responsible disclosure

Found a hole. Tell us.

report

security@leaksweep.net
  • >Acknowledged by a person within one business day.
  • >No legal action against good-faith research that follows this page.
  • >We tell you what we fixed and when, and credit you if you want it.
  • >Please give us 90 days, or until the fix ships, before publishing.

ground rules

  • >Test against your own account. Do not access, change, or delete another user's data.
  • >No denial of service, no volume attacks, no social engineering of staff.
  • >If you reach data that is not yours, stop, do not save it, and tell us.
  • >Mask any credential in your report. We will ask for detail if we need it.

// in scope: leaksweep.net, the console, the sweep API.
// out of scope: our hosting, model, and payment providers. report to them.

what we do not have yet

The honest list.

Leaksweep Inc. is a pre-seed company with a team of 7. If your procurement process needs an item below, we are not the right vendor today, and we would rather you hear that from us.

  • -No SOC 2 report. We are a pre-seed company of seven. A Type I audit is planned once revenue supports it, and we will not imply one before it exists.
  • -No ISO 27001 certification.
  • -No third-party penetration test on file yet. It is planned ahead of the SOC 2 work, and the summary will be published here.
  • -No paid bug bounty. Reports are credited, with permission, and answered by an engineer.
  • -No SSO or SCIM. Sign-in is email and password, GitHub, or Google.
  • -Rate limiting is per instance, not a distributed limiter.

Related: security-first care, your data, privacy policy.

$ leaksweep init

Sweep your code before someone else does.

Three sweeps a month on the free tier. No card. Secrets masked from the first byte.