Skip to content

PATCH-v0.1.1.md

Latest commit

 

History

History
49 lines (41 loc) · 3.61 KB

PATCH-v0.1.1.md

File metadata and controls

49 lines (41 loc) · 3.61 KB

ai-rsk v0.1.1 - Patch Notes

Bug Fixes

EVAL_DYNAMIC false positive fix

  • Problem: Pattern eval\s*\( matched the word "retrieval" followed by ( in comments and prose (e.g., // Store in sessionStorage for retrieval (no actual generation yet))
  • Fix: Added \b word boundary before eval on all 3 eval patterns to ensure only standalone eval( is matched, not substrings like retrieval(
  • Problem: Generic exec() pattern caused false positives on JavaScript methods named exec (function declarations, comments, user-defined functions)
  • Fix: Removed the generic exec() pattern from eval-dynamic.yaml (JS/TS only). Created a new dedicated rule EXEC_DYNAMIC_PYTHON (exec-dynamic-python.yaml) targeting only *.py files with ^\s*exec\s*\( anchor to avoid matching method calls like db.exec()
  • Fix: Improved promisify pattern to catch execAsync = ...promisify(child_process.exec) with variable name suffixes
  • Verified on: ai-alixia, alixia-gen (geminiService.ts false positive confirmed eliminated), SoS, Majordome, Blinko, Ai-Ops, agentic-workflow, AI-Context-Engineering, Kage, searchly

Cookie negation per-file fix

  • Problem: MISSING_HTTPONLY, MISSING_SECURE_COOKIE, MISSING_SAMESITE used project-wide agnostic negation. If ONE file had httpOnly: true, the rule was satisfied for ALL files in the project, hiding cookies without flags in other files.
  • Fix: Added agnostic_negation field to Rule struct (default: true for backward compatibility). Set to false on the 3 cookie rules. When false, negation is checked per-file only.
  • Verified on: test with 2 files (auth-safe.js with httpOnly + tracking-vulnerable.js without) - only tracking-vulnerable.js triggers, auth-safe.js correctly passes.

Exclusion improvements

  • Added: bundle directory to EXCLUDED_DIRS in rules.rs, runner.rs, analyze.rs (eliminates false positives on bundled/compiled code like Majordome's apps/backend/bundle/main.js)
  • Added: *.min.js, *.min.css, *.min.mjs file exclusion in is_excluded() (eliminates false positives on minified third-party libraries like echarts.min.js, markmap.min.js in Blinko)

New Rule

  • EXEC_DYNAMIC_PYTHON (CWE-95, BLOCK) - Detects Python exec() and eval() with dynamic input, including f-string variants. Targets *.py files only.

Stats

  • Rules: 31 -> 32 (1 new Python rule)
  • Tests: 162 passing
  • Clippy: 0 warnings

Files Changed

File Change
rules/eval-dynamic.yaml Removed generic exec pattern, removed Python, added \b word boundary
rules/exec-dynamic-python.yaml NEW - Python-only exec/eval detection
rules/missing-httponly.yaml Added agnostic_negation: false
rules/missing-secure-cookie.yaml Added agnostic_negation: false
rules/missing-samesite.yaml Added agnostic_negation: false
src/rules.rs Added agnostic_negation field, per-file negation logic, bundle dir, min.js exclusion
src/runner.rs Added bundle to SCAN_EXCLUDED_DIRS
src/analyze.rs Added bundle to EXCLUDED_DIRS
src/embedded_rules.rs Added exec-dynamic-python.yaml include
tests/fixtures/vulnerable/eval.js NEW - eval test fixture
tests/fixtures/safe/eval.js NEW - safe eval test fixture

Validation

Tested on 10 real repositories:

  • ai-alixia, SoS, Ai-Ops, Majordome, agentic-workflow, AI-Context-Engineering, Kage, Blinko, alixia-gen, searchly
  • Zero false positives on EVAL_DYNAMIC across all 10 repos (1 residual on Blinko: tex-svg-full.js third-party lib, configurable via exclude)
  • Cookie per-file negation working correctly on ai-alixia (server-local.ts real positive detected)