Repository navigation
fix(tinyredact): redact cards next to other digits and quoted or prefixed credentials - #308
Merged
M3gA-Mind merged 1 commit intoOct 22, 2026
Conversation
…ixed credentials A separated card run that failed Luhn as a whole (a card followed by a CVV or preceded by a quantity) was dropped, because the greedy card regex cannot give digits back. Such a run is now retried on windows of its digit groups; a window must pass Luhn and be corroborated. The credential patterns allowed no quote between the key and its separator, so JSON-quoted credentials in prose survived, and their leading word boundary never fired after an underscore, so DB_PASSWORD=, *_SECRET= and *_TOKEN= survived. Both now match; the trailing word boundary still keeps token_count, max_tokens and tokenizer clean.
|
Warning Review limit reached
This review includes 5 billable files and costs up to $1.25. Or wait 26 minutes for your next included review. View limit details
Comment |
Tiny Sweeper reviewReviewed head: Verdict: Ready for maintainer review · Priority: none Diagnostics and supporting detailsFull summaryPull request purpose (title): fix(tinyredact): redact cards next to other digits and quoted or prefixed credentials. Tiny Sweeper completed its review; deterministic results follow. Review snapshot
Completeness: Complete What changedCard detection now routes every `CC_RE` match through a new `card_span` helper. A run that passes Luhn is still taken as the card (under LuhnOnly, when separated, or when a bare run is corroborated). A separated run that fails Luhn as a whole is retried on windows of its contiguous digit groups, 13-19 digits, longest first then leftmost, and a window must pass Luhn and be corroborated by `plausible_card_number` or `cc_keyword_near` under either gate. Bare runs are a single group and are never split, so the epoch-millisecond protection is unchanged. Credential patterns in `patterns.rs` now accept an optional quote between the key and its separator for `api_key`, the `password`/`secret`/`token` group, and the OAuth-names group, and the broad credential pattern starts with `(?:\b|_)` while keeping the trailing `\b`. The description notes two accepted side effects: `has_likely_secret` will also refuse lookalikes such as `is_secret: false`, and a prefix match starts at the `_`, so `DB_PASSWORD=x` renders as `DB[REDACTED]`. Not included: a `secret_access_key` arm and the fail-closed `dedupe_overlaps` merge. Features
Tests
FindingsNo active actionable findings. Before mergeNone. Agent review detailscritique
security
tests
commits
description
Evidence and run details
|
![tinysweeper[bot]](https://avatars.githubusercontent.com/in/4526585?s=60&v=4){kind=small}
tinysweeper
Bot
added
the
priority: p3
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Three content-redaction gaps let real credentials through
redact_pii*/sanitize_text*, under both card policies.Card next to other digits.
CC_REis greedy, socard 4111 1111 1111 1111 123matches all 19 digits. That run fails Luhn, andfind_iternever retries a shorter one, so the card survived. A leading number (qty 2 4111 1111 1111 1111) fails the same way. The regex crate has no lookaround to fix this in the pattern.plausible_card_number) or a card keyword nearby (cc_keyword_near). This applies under either gate, because splitting a run multiplies the chance of a Luhn pass by luck.card_spanhelper. TheBareCardGatedoc now describes the retry.JSON-quoted credentials in prose. The
api_keypattern and thepassword/secret/tokenpattern allowed no quote between the key and its separator. As a result,{"password": "hunter2", "api_key":"abcd1234"}inside text was left as is. The OAuth-names pattern had the same gap. All three now accept an optional quote there.DB_PASSWORD=/*_SECRET=/*_TOKEN=. The leading\bnever fires after_, so prefixed environment-style assignments survived. The pattern now starts with(?:\b|_). The trailing\bis kept on purpose: it is what leavestoken_count=5,max_tokens: 10,tokenizer=x,password_reset_sent: trueandsecret_santa=…alone.Related issue
None. Found in an audit for tinyhumansai/openhuman.
API or behavior changes
No API change. Behaviour only ever adds redactions:
_-prefixed credential assignments are redacted. The prefix match starts at the_, soDB_PASSWORD=xbecomesDB[REDACTED]. That is cosmetic; the value is gone.has_likely_secret, which openhuman uses to refuse flow-memory writes. The intended new refusal is{"token":"abc"}. Reading the regex, it will also refuse lookalikes whose value follows a credential word directly, e.g.is_secret: false,has_password: true,max_token=5. Those false positives are accepted rather than weakening the pattern. Making the match depend on the value is the follow-up if they show up in practice.secret_access_key(AWS_SECRET_ACCESS_KEY=,STRIPE_SECRET_KEY=), and the fail-closed merge of partially overlapping hits indedupe_overlaps. Both were left out to keep this change to the three reported gaps.Validation
cargo fmt --all -- --checkwas not run;rustfmt --edition 2024was run on each touched file. CI runs the full check.cargo clippy --all-targets --all-features -- -D warningswas not run locally; CI runs it.cargo build --all-targets --all-featureswas not run locally; CI runs it.cargo test --all-featureswas not run locally. Only the six tests below were run, in one targeted invocation; CI runs the full suite.cargo test -p tinyredact --lib -- <the six tests below>:pii/mod.rsandsanitize/patterns.rsreset tomain(tests kept): FAILED, 2 passed (the two guards), 4 failed:credit_card_followed_by_space_and_digits_redacted:card survivedunder both policies;credit_card_after_a_leading_number_is_redacted_and_the_number_kept:card survived: "qty 2 4111 1111 1111 1111";sanitize_text_redacts_json_quoted_credentials_in_prose:hunter2andabcd1234survived under both policies;sanitize_text_redacts_prefixed_env_style_credentials: all three values survived under both policies.Tests
a_separated_digit_list_with_no_card_is_left_alone:ids 123 1000 2000 1000 0002. The whole run fails Luhn, and its only Luhn-valid window has no network IIN and no keyword, so it stays unredacted under both gates.sanitize_text_leaves_credential_lookalike_names_alone:token_count,max_tokens,tokenizer,password_reset_sentandsecret_santaare unchanged.Documentation
The
BareCardGatedoc and thecard_spandoc describe the window retry. No external docs cover these patterns.Checklist
#[allow(...)],#[ignore], or relaxed lints.envcontents in the diff or the description