Skip to content

Redact sensitive data from every bug report, before upload

Updated

Midge redacts sensitive data from bug reports automatically. Passwords, card numbers, tokens and cookies are redacted in your browser before anything leaves your device. The same rules cover headers, URLs, request and response bodies, console messages, instant replays and the user action log, and every removed value is replaced with [REDACTED] so the reader knows something was there.

The guiding rule is simple: when in doubt, hide it. Removing a harmless value is an acceptable cost. Leaking a session token is not. This page lists every rule in plain language, then explains what redaction does not do. The short version lives on our security page.

The rules

Headers

These headers are always redacted by name: Authorization, Proxy-Authorization, Cookie, Set-Cookie, X-API-Key, X-Auth-Token, X-CSRF-Token and X-XSRF-Token.

On top of that, any header whose name contains token, secret, key, session, auth, signature or password is redacted, in any letter case. So X-Shopify-Access-Token and X-Session-Id are covered without anyone listing them.

URLs

In request URLs, page changes and URLs that appear in console text, the values of these parameters are removed: token, access_token, refresh_token, id_token, code, key, api_key, apikey, signature, sig, password, pass, secret, session, sessionid, auth and jwt. This applies to the query string and to the fragment after #, which is where many OAuth flows put tokens. A password written into a URL, as in https://user:pass@host, is removed too. Any parameter whose value matches one of the patterns below is redacted regardless of its name.

Request and response bodies

Midge scans JSON, URL-encoded forms and multipart forms, including nested objects. Fields with these names are redacted: password, passwd, pwd, pass, secret, token, otp, pin, cvv, cvc, card, cardNumber, pan, iban, ssn, apiKey, privateKey, client_secret, refresh_token and access_token. Matching ignores the difference between camelCase and snake_case, and names that end in one of the core words, such as newPassword or reset_token, match as well.

A login request looks like this after redaction:

{
  "email": "j***@acme.com",
  "password": "[REDACTED]",
  "device": { "name": "Work laptop", "pushToken": "[REDACTED]" }
}

Values that look secret, wherever they appear

Some secrets are recognizable by their shape alone. Midge removes these from bodies, headers, URLs, console text and page text in replays, whatever the field is called:

  • JSON Web Tokens: three base64url parts separated by dots, starting with eyJ.
  • Card numbers: 13 to 19 digits, with or without spaces or dashes, that pass the Luhn checksum. Order numbers that fail the check are left alone.
  • API keys in known formats: Stripe (sk_live_, sk_test_, rk_live_), AWS (AKIA), GitHub (ghp_, gho_, github_pat_), Slack (xoxb-, xoxp-), Google (AIza), and OpenAI and Anthropic style keys (sk-, sk-ant-).
  • Bearer tokens: the value after Bearer.
  • Private keys: any block between -----BEGIN ... PRIVATE KEY----- and its end line.

Emails and phone numbers

These are masked, not removed, so the report still shows which account was involved. [email protected] becomes j***@acme.com. International numbers that start with + keep the country code, the first digit and the last two digits: +44 7700 900123 becomes +44 7*** ****23.

What you type

Typed text is never recorded. The action log only says that you typed in a field, for example "Typed in input[name=email]". Text in click targets goes through the same pattern scan as everything else.

Video, screenshots and replays

  • In tab recordings, password fields, card fields (autocomplete="cc-...") and any element marked with data-midge-blur are blurred while you record, so they never reach the video.
  • In screenshots, you can blur any area by hand.
  • In instant replay, all input values are masked and page text goes through the pattern rules above.

Your own rules

Every app has its own secrets: a tenantKey, an X-Partner-Id. Workspace admins can add field names and header names, and the extension picks up the list when you sign in. You can add rules, but you cannot remove the built-in ones. There is no off switch and no allowlist, by design.

What redaction does not do

Being clear about the limits matters as much as the rules:

  • It is rule-based. It removes what the rules describe. A secret with an unusual name and an unrecognizable format can slip through. Add a workspace rule when you find one.
  • Random-looking strings are kept. Midge does not guess based on entropy, because that would also wipe request IDs, hashes and record IDs that developers need to debug.
  • Window and full-screen recordings are not blurred. Only tab recordings blur fields automatically. Before a window or screen recording, Midge reminds you to close anything sensitive.
  • Page text in a tab video is not scanned. If a token is printed on the page, it is visible in the video. Instant replay is different, because it works on page text rather than pixels.

Try the rules on your own file

The free HAR sanitizer uses exactly the same redaction code as the extension. Drop in a HAR file exported from DevTools and see what would be removed, all in your browser. For the manual route, read how to remove sensitive data from a HAR file. Redaction also applies to everything in the network logs and console logs of a report.

Midge is free: unlimited recordings, videos up to 15 minutes, and no watermark.

Frequently asked questions

Where does redaction happen?
In your browser, on your device, while Midge prepares the report. Midge servers only ever receive the redacted version, so the raw values are never stored.
Can I turn redaction off for a site I trust?
No. There is no off switch and no allowlist. Teams can add more field and header names to redact, but they cannot remove the built-in rules.
Does Midge catch every secret?
No tool can promise that. Midge follows a fixed, tested set of rules. A secret with an unusual name and format may not be recognized, which is why teams can add their own field and header names.
Why are emails masked instead of removed?
A developer often needs to know which account hit the bug. Half-masking, such as j***@acme.com, keeps that signal without exposing the full address.
Is my whole screen blurred when I record a window?
No. Window and full-screen recordings are not blurred automatically. Midge warns you before such a recording starts so you can close anything sensitive.