Repository navigation
Releases
@link-foundation/sensitive-data-sanitizer 0.23.0
- c6c94b2: Add
identifiersandtransliterateflags for known values, settable per entry or ascompileKnowndefaults, to find listed names inside code identifiers and in common Latin spellings of Cyrillic words. Match name parts split by one line break with an optional comment or quote prefix. Report a listed part nested in a listed full name once, require the surname or its own entry for a given name with a patronymic, and match a given name with the surname initial. Compile and run only the known-value expressions whose literal anchors occur in the text, and addcompileKnown(..., { warm: true }), so the first call with 2,500 entries takes about 1 ms instead of 8 s.
Related Pull Request: #191
@link-foundation/sensitive-data-sanitizer 0.22.5
- 8ee08c3: Preserve Authorization placeholders and trailing literal backticks in Markdown code spans with longer delimiters. Keep embedded backticks and literal credential suffixes sensitive. Stabilize JavaScriptCore lookbehind regression checks with warm-up and median sampling while retaining detection of quadratic growth on noisy CI runners.
Related Pull Request: #185
@link-foundation/sensitive-data-sanitizer 0.22.4
-
Sanitizing text that holds a long letter run and any character above U+00FF, such as an emoji,
—or axx-…789mask, no longer fails withERR_SANITIZATION(#172). With theuflag, V8 compiled the\p{L}/[^\s]loops of the credential-key, camelCase-key, e-mail, mail display-name, URL, document-link, token and projection expressions as loops with one backtracking entry per character on such strings, so a 4 MiB run overflowed its regular-expression stack. Those loops now list the BMP members of their class and allow at most 256 astral members per run. They match the same text and stay flat on runs of any length. A long value after a credential keyword (password=orapi_key:followed by 1 MiB) is now scanned in about 1 s rather than 13 s (#173), because the vendored Gitleaks rules scan a view in which the inside of each long base64 run is replaced by a fixed filler, with offsets mapped back. SetSENSITIVE_DATA_SANITIZER_DEBUG=1to print the stack of an unexpected internal failure to stderr.Streaming no longer turns quadratic after a line that only looks like the start of a netrc entry, such as
default = ["macro"],default LocaleContext …ormachine learning is fun(#176). The stream tracks themachine <name>ordefault,login <user>,password <value>token sequence and releases held text as soon as a token breaks it, so each token costs constant time. Under Node.js,sanitizeFileToFileon 8,000 log lines afterdefault = ["macro"]took 6.5 s and now takes 0.44 s, the same as without that line. Genuine multi-line entries are still held and redacted, including an entry whosedefaultormachinekeyword ends a line of other text. In a Markdown code span, anAuthorization:value now ends at the span's closing backtick (#177):`Authorization: token <value>`,`Authorization: Bearer $TOKEN`and`Authorization: Bearer ${TOKEN}`stay readable, and a real credential is masked as`Authorization: token [REDACTED]`with its closing backtick kept.Preserve NUL separators and subsequent log fields when redacting credentials with
--allow-binary, including whole-input and streaming paths. Contextual credential values also stop at NUL. Uniformly NUL-interleaved text remains detectable without joining ordinary NUL-delimited fields to a credential.
Related Pull Request: #183
@link-foundation/sensitive-data-sanitizer 0.22.3
- e0864bb: Sanitizing text with one long single-line letter run, such as a base64 blob embedded in a JSONL log, no longer fails with
ERR_SANITIZATIONon Node.js (#170). The camelCase credential key scan (clientSecret: …) matches the words before the credential word with one character class, which accepts the same keys without a backtracking entry per capital letter, so an 8 MiB or longerAAAA…/AbAb…run no longer overflows V8's regular-expression stack. Separated credential keys (spring.datasource.password,DB_PASSWORD_FILE) and obfuscated e-mail addresses (jane.roe (at) example (dot) org) accept at most 16 separated words or dotted parts on each side, soA_A_…anda.a.a.…runs no longer overflow either, and aa-a-a-…line no longer takes quadratic time (256 KiB took about 100 seconds; it now takes about 40 ms). Wrapped base64 runs (base64 -w, PEM bodies) and percent-encoded runs (%41%41…) are matched line by line and escape by escape, andU+0041 U+0042 …and CSS\41 \42 …escape runs unit by unit, with the same results, so long runs of those no longer overflow the stack either. A WordPressdefine('AUTH_KEY', '…')value repeats a group only per escape, so a long value is found instead of failing.
Related Pull Request: #171
@link-foundation/sensitive-data-sanitizer 0.22.2
- e1056c9: Re-sanitizing text that already holds masked credentials (
ghp…b4N,Bearer 012…567) costs close to plain text again (#168). The detection view keeps thehive-maskellipsis instead of turning it into..., so a masked log no longer scans the original and the normalized text separately; plain ASCII runs are copied into the view in bulk; a clean result is not inspected a second time; and open-ended credential spans are bounded by findings at the same start instead of by every other finding. 6000 lines with a masked GitHub token went from about 300 KB/s to 1.5–2.9 MB/s, and masked bearer headers from about 46 KB/s to 0.6–0.9 MB/s.
Related Pull Request: #169
@link-foundation/sensitive-data-sanitizer 0.22.1
-
Recover complete unpadded wrapped base64 text before following prose so encoded JSON retains its structure, fields and wrapping during sanitization.
Label the diagnosis and community acronyms
СПИД,СДВГ,ПТСР,PTSDandЛГБТ/LGBTQ+in any letter case and redact a hyphenated compound around a special-category term (вич-инфекция,женщина-врач) as one attribute. Public findings now report one personal finding per span and type (#164). Structured JSON sanitization inspects each distinct key and string value once per call, and pattern scans no longer clone their expressions, cutting a 32-field log record from about 73 ms to about 15–18 ms (#165).
Related Pull Request: #167
@link-foundation/sensitive-data-sanitizer 0.22.0
-
dd1abdf: Fix strict personal-attribute coverage and type precedence, reuse vocabulary compilation, keep reviewed public technology names readable, and make placeholder prechecks linear and reusable. Recognize redacted PEM bodies next to end markers without repeatedly failing publication verification.
Add explicitly enabled, bounded read-only checks for twelve credential providers and an isolated local Betterleaks detector with live validation disabled. Add documentation navigation, workflow guidance and an evidence-based parity audit with explicit outstanding requirements.
Related Pull Request: #147
@link-foundation/sensitive-data-sanitizer 0.21.8
@link-foundation/sensitive-data-sanitizer 0.21.7
@link-foundation/sensitive-data-sanitizer 0.21.6
- cbd6b3b: Preserve complete credential placeholders and public configuration values before Markdown backticks, including the exact
HIVE_MIND_CLAUDE_AUTH_AUTHORITY=conventionalopt-out. Keep masking actual HTML/script and keyword-prefixed credentials, with exact known secrets taking priority.
Related Pull Request: #158