Kingfisher v2.11.0

[v2.11.0]

• Handled malformed Git tree modes without panics, guarded staged tree ancestry, normalized legacy index modes, and avoided revisiting reachable commits when including unreachable objects.

• Propagated strict archive member errors, removed partial ZIP/TAR files, preserved TAR/ZIP bytes when no output directory is supplied, and bounded best-effort SQLite schema listing with skipped-table warnings.

• Pruned stale compiled-cache temporary files, preserved existing entries when replacement fails, reported cache rejection reasons once per process, and bound line-filter cache keys to source identity without cloning patterns on hits.

• Validated Python archive/content options eagerly, excluded case variants of .git, rejected unrepresentable Git hour bounds, and clarified extraction budgets and format-specific entry accounting.

• Preserved separately encoded secrets in nested Base64 siblings by scoping CLI and SDK containment checks to each decoded buffer.

• Refreshed the Betterleaks catalog to v2.0.0-rc.1,
  adding Bitbucket Data Center HTTP access tokens, Cloudflare cfut_/cfat_ tokens and their
  account-ID helper, and fixing Tableau personal access tokens beginning with + or /.
  The legacy Cloudflare rule is now betterleaks.cloudflare-api-key.1; the family selector
  continues to cover both formats. Updated expression-helper support preserves upstream
  validation checks. Built-in coverage is now 488 rules (462 Betterleaks and 26 Veles),
  with 255 supporting validation. Regenerated the bundle, provenance, source archive, and docs.

• Preserved every repeated TAR member during extraction and made SDK compressed-stream budget exhaustion fail before returning partial archive members.

• Reused compiled rule caches across compatible deployments with exact engine build identity and automatic recompilation on native CPU/version rejection. #538

• Enabled Python SDK caching by default with Rules(cache_dir=...) and Rules(cache=False) controls.

• Fixed SDK lazy-span scan slowdowns (up to 50× faster) without changing findings. #536

• Fixed dense-finding scan slowdowns in regex confirmation, catalog/component matching, URI overlap, and inline-ignore checks. #537

• Improved dense scans with long lines, chained components, Base64 URIs, and HTML/CSS; fixed inconsistent source locations.

• Added regression benchmarks: up to 144× faster SDK and 74× faster CLI on the issue reproducer, with unchanged finding counts and comparable or improved repository and Git-history scan times.

• Added composable Python filesystem and Git-history enumeration through ScanInput, filesystem(), git_history(), and Scanner.scan_inputs(). Callers can supply their own Python iterators or use native enumeration without changing existing SDK signatures or behavior.

• Added explicit Python archive expansion with shared CLI extractors, nested-depth and per-root budgets, cooperative controls, and source/Git metadata retained with each findings group.

• Added commented Python examples for filesystem enumeration, native versus Python-managed Git history, and archive scanning from files or bytes; documented composition, limits, and validation grouping.

• Preserved invalid UTF-8 secrets during HTML/CSS context filtering in the CLI and SDK.

• Added opt-in Python DetectionScanner and Rust detection options for shared CLI confirmation/component-window semantics, credential-URI fallback and secret containment suppression, bounded two-level Base64 decoding, inline-ignore and HTML/CSS context filtering. Existing SDK signatures and defaults remain unchanged.

• Isolated new CLI helper APIs behind explicitly unstable features, kept optimized confirmation state opaque, and preserved exhaustive matches of the existing embedding confirmation enum.

• Added CLI/SDK parity, nested Base64 limit, full-match component-anchor, and context-phase interruption regressions; interrupted scans return no partial findings or dedup commits.

• Added composable expand_content() for shared SQLite SQL and Python bytecode string extraction from files, historical blobs and archive members, with source provenance, per-input budgets and controls retained.

• Added native git_inputs() with GitScope history baselines, inclusive branch roots, time windows, snapshots, net diffs, staged index content and unreachable object coverage; GitInput retains multiple scoped occurrences, author/committer identities and times, parents and messages without modifying repositories.

• Added commented detection, SQLite/bytecode and Git-scope examples, contract/regression tests and updated SDK/Rust documentation.

• Added frozen Python DetectionPolicy configuration, direct native finding properties and structured result conversion, and scanning of borrowed Python bytes.

• Improved dense multiline credential scanning by reusing confirmed matches while preserving capture values and offsets.

• Shared native Git scope enumeration with the Rust library, skipped unchanged subtrees, shared commit metadata, and added blob/preparation limits, non-UTF-8 path provenance and explicit missing-object handling.

• Enforced nested extraction budgets while writing, hardened SQLite parsing, supported temporary-directory selection and reported malformed-content fallback diagnostics. Corrected Python 3.11+ marshal code-object parsing and handled Python 3.14 slice constants.

• Added cancellable validation/revocation batch deadlines, Ctrl-C handling and safe revocation error categories.

• Hardened compiled cache ownership/permissions and payload integrity, removed the shared temporary fallback, and preserved exact native engine build identity.

• Bumped kingfisher-core to 1.0.3, kingfisher-rules to 1.2.0 and kingfisher-scanner and Python SDK to 1.3.0.