fix(redact): strip complete CSI sequences before prefix-mask scan - #81083
Closed
Enough1122 wants to merge 2 commits into
Closed
fix(redact): strip complete CSI sequences before prefix-mask scan#81083Enough1122 wants to merge 2 commits into
Enough1122 wants to merge 2 commits into
Conversation
…usResearch#81012) Bare-ESC stripping left CSI terminator bytes (`m` / `K` / ...) glued to the token head in the stripped copy, which defeated the `(?<![A-Za-z0-9_-])` prefix-regex lookbehind and let whole tokens leak through (`\x1b[32msk-...\x1b[0m`). Fix: in `_mask_control_split_tokens`, eat COMPLETE CSI sequences (ESC + `[` + params + terminator letter) instead of just the bare ESC byte. Track eaten positions so the body-safe check accepts the CSI bytes when they fall inside an original mask span. Also handles the second gap from the same report: a split that crosses \n AND has a self-matching fragment on one side used to skip the cross-line mask entirely (the legacy line-boundary guard). When that happens, fall back to per-line masking so the prefix-bearing line is still masked instead of the whole token leaking. Regression tests cover the real leak shape (`\x1b[32mtok\x1b[0m`, no space between the CSI terminator and the token), the same shape with a space, and the ESC + newline split case. Fixes NousResearch#81012
Enough1122
force-pushed
the
fix/81012-redact-csi-prefix-masking
branch
from
August 7, 2026 14:09
c448f81 to
8f07f93
Compare
The locally re-declared control-char regex had its \uXXXX escapes mangled into literals, forming an unintended \x7f-\u200f range that matched Latin-1 high bytes (é, £, ...) as "control" chars. Reuse the module-level _CONTROL_CHARS_RE (already correct) for the shadow-copy strip and body-safe check so the two cannot drift.
alt-glitch
added
type/security
Contributor
Author
|
Duplicate of #81060 (same issue). Closing in favor of the earlier PR to avoid double-fixing the same issue. |
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
2 participants
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.
Fixes #81012
_mask_control_split_tokensstripped only the bare ESC byte, leaving the CSI terminator letter (m/K/ ...) glued to the token head in the shadow copy. The prefix-regex lookbehind(?<![A-Za-z0-9_-])then failed and the whole token leaked verbatim through logs (\x1b[32msk-...\x1b[0m).Fix: eat COMPLETE CSI sequences (ESC +
[+ params + terminator) before the prefix scan, and track the eaten positions so the body-safe check accepts them when they fall inside an original mask span.Also closes the second gap from the same report: a split that crosses \n AND has a self-matching prefix on one side used to skip the cross-line mask (legacy line-boundary guard). When that happens, fall back to per-line masking so the prefix-bearing line is still masked instead of the whole token leaking.
Regression tests cover the real leak shape (
\x1b[32mtok\x1b[0m, no space), the same shape with a space, and the ESC + newline split case.