Repository navigation
TorusGuard v2.1.1: Taint Engine, Polyglot AST Walker & 88 Rules Expansion
TorusGuard v2.1.1: Taint Engine, Polyglot AST Walker & 88 Rules Expansion
Release Version: v2.1.1
Release Date: September 29, 2026
Package Manager: npm install -g torusguard@2.1.1
NPM Package: torusguard@2.1.1
GitHub Packages: @githubmofo/torusguard@2.1.1
Binary Engine: Zero-dependency Go 1.25 Single Binary (cmd/torusguard) with Python polyglot core runtime
Executive Summary
TorusGuard v2.1.1 delivers a major static analysis enhancement, introducing the Taint Analysis Engine, Polyglot Tree-Sitter AST Walker, Cross-File Interprocedural Dataflow, and 7-Signal Calibrated Confidence Scoring. It expands the canonical rule catalog to 88 rules across 22 architectural families by adding deep detection for unvalidated URL redirects (TG-INPUT-007) and insecure object deserialization (TG-INPUT-008).
This release also hardens type safety across finding evaluation scripts, ensuring seamless developer experience and zero type warnings in modern AI IDEs.
Key Highlights & Innovations
┌─────────────────────────────────────────────────────────────────────────────┐
│ TORUSGUARD v2.1.1 MILESTONES │
└─────────────────────────────────────────────────────────────────────────────┘
1. Taint Analysis Engine ──► Source-to-sink dataflow tracking across ASTs
2. Polyglot AST Walker ──► Tree-sitter powered parser (Python, TS/JS, Go)
3. Cross-File Interprocedural ──► Cross-module call graph and import resolution
4. 88 Rules Across 22 Families──► Added TG-INPUT-007 and TG-INPUT-008
5. 7-Signal Confidence Scorer ──► Empirical evidence-chain calibrated scoring (0–100)
6. Incremental Scan Cache ──► SHA-256 mtime hashing for sub-second re-audits
7. Finding Scorer Type Safety ──► Zero Pylance/Pyright typing warnings
8. Master 7-Stage Pipeline ──► torusguard full closed-loop governance command
9. Tri-Mode Validation Suite ──► 6 unseen stacks tested with 100% pass across CLI/Chat/MCP
Detailed Feature Breakdown
1. Taint Analysis Engine (core/taint.py, core/taint_rules.py, core/taint_graph.py)
- Tracks untrusted inputs (sources:
request.args,req.query,params,os.environ,body) across intermediate variable assignments to critical security execution sinks (eval,subprocess,db.query,fs.readFile,res.redirect). - Recognizes sanitizers (
int(),escape(),sanitize(),DOMPurify.sanitize,encodeURIComponent) to eliminate false positives. - Constructs directed acyclic taint graphs capturing propagation steps, path depth, and line-level coordinates.
2. Polyglot Tree-Sitter Parser & AST Walker (core/parser.py, core/ast_walker.py, core/symbol_table.py)
- Standardizes CST and AST node traversal across Python, TypeScript, JavaScript, and Go.
- Maintains lexical scopes and symbol tables tracking identifiers, function definitions, parameters, and variable assignments.
3. Cross-File Interprocedural Dataflow (core/cross_file_taint.py, core/call_graph.py, core/import_resolver.py)
- Resolves cross-file imports (relative paths, package paths, index barrel exports).
- Connects callers and callees into a project-wide call graph, allowing taint flow to propagate across module boundaries.
4. 88 Canonical Rules Across 22 Architectural Families
- Expanded the
TG-INPUTfamily from 6 to 8 rules:TG-INPUT-007: Unvalidated URL Redirects / Open Redirection (rules/TG-INPUT-007-unvalidated-redirect.md).TG-INPUT-008: Insecure Deserialization / Object Injection (rules/TG-INPUT-008-insecure-deserialization.md).
- Loaded and validated via
core/rules_registry.pyand categorized across 22 root-cause clusters incore/clustering.py.
5. 7-Signal Calibrated Confidence Scorer (core/confidence.py, .torusguard/scripts/finding_scorer.py)
- Calibrates 0–100 confidence bands using 7 weighted empirical signals:
- Rule severity baseline (Critical=90, High=75, Medium=50, Low=25, Informational=10)
- Taint path confirmation (+15)
- Taint depth decay
- Sanitizer presence suppression (-35)
- Framework context match (+5)
- Multi-line evidence quality (+5)
- Test/Mock fixture noise suppression (-30) and documentation suppression (-25)
- Persistent memory boost/suppression (-30 to +20)
6. Incremental Hash Cache & Parallel Scanning (core/incremental.py, core/parallel.py, core/watcher.py)
- Persists SHA-256 and modification time hashes at
.torusguard/cache/audit_cache.jsonto skip unchanged files. - Distributes directory chunk auditing across CPU cores using a process worker pool.
- Includes file-system watcher with 300ms debounce for continuous background monitoring.
7. Master 7-Stage Pipeline (torusguard full)
- Coordinates the complete closed-loop security governance lifecycle:
- Stage 0:
init(Stack discovery & rule profiling) - Stage 1:
authorize(Cryptographic target scope validation) - Stage 2:
audit(Polyglot AST & multi-modal vision scan) - Stage 3:
verify(Finding confidence calibration & line fingerprinting) - Stage 7:
report(Living ledger synchronization, SARIF v2.1.0 & HTML export)
- Stage 0:
8. Stream-Isolated MCP Server & Unseen Repositories Validation
- Redirected stdout to stderr inside
RunMCPServer()incmd/torusguard/mcp.go, ensuring 100% pure JSON-RPC 2.0 stdio compliance. - Validated against 6 unseen ecosystems (SvelteKit 2 + Bun, FastAPI pgvector RAG, Cloud-Native DevOps Docker & Git Mine, Multi-Modal Vision OCR on PNG/JPG/WebP, Kotlin Ktor + Exposed, and Laravel 11 + Vue 3) with full automated sandboxed cleanup.
Installation & Upgrade
Global Installation via npm
npm install -g torusguard@2.1.1Direct Usage via npx
npx torusguard@2.1.1 auditBuild from Source
git clone https://github.com/githubmofo/TorusGuard.git
cd TorusGuard
git checkout v2.1.1
go build -o torusguard ./cmd/torusguard