Scan Modes

Tameshi offers three complementary analysis modes that can be used independently or combined.

Overview

Mode	Speed	Precision	Works on Incomplete Code	Use Case
Source	<1s	Medium	✅ Yes	Real-time feedback, CI/CD
IR	~3-5s	High	⚠️ Requires valid syntax	Pre-deployment audits
Combined	~10-15s	Higher	⚠️ Depends on LLM input	Comprehensive reviews

Source-Level Scanning

Fast analysis (<1s) on incomplete, under-construction code.

Key Advantage: Works on Invalid Code

Critical for active development:

• Scans syntactically invalid Solidity
• No compilation required
• Tree-sitter parser is error-tolerant
• Get feedback while typing, not just on complete code

Features

• Security feedback in <1 second
• Scans code as you type in VSCode
• Runs on every save automatically
• Catches vulnerabilities in incomplete functions

How It Works

Source scanners use tree-sitter to parse Solidity into an error-tolerant AST, then match vulnerability patterns. Works even with syntax errors. Perfect for active development.

14 source-level scanners across all vulnerability categories.

When to Use

Good for:

• Real-time feedback during coding
• CI/CD pipelines where speed is critical
• Pre-commit hooks
• Incomplete code under construction

Example

tameshi scan run -i MyContract.sol  # <1s even on invalid syntax

VSCode: Automatically scans on save (default behavior).

---

IR-Level Scanning

Precise semantic analysis through ThalIR intermediate representation.

Features

• Detects vulnerabilities that source scanning misses
• Analyzes cross-function vulnerability chains
• Provides high-confidence findings with low false positives
• Performs data flow and control flow analysis

How It Works

Tameshi transforms Solidity to ThalIR (intermediate representation), then runs scanners on the IR. This enables semantic analysis: understanding what the code does, not just how it's written.

ThalIR provides:

• SSA (Static Single Assignment) form for precise data flow
• Control flow graphs for path analysis
• Type information for safety checks
• Normalized representation that's easier to analyze

When to Use

Good for:

• Pre-deployment security audits
• Critical smart contracts handling value
• Contracts with complex logic
• When precision matters more than speed

Trade-offs:

• Approximately 10x slower than source scanning
• Requires Solidity compilation to succeed
• More resource-intensive

Example

# IR scanners run automatically with deterministic suite
tameshi scan run -i MyContract.sol --suite deterministic

The transformation to IR happens transparently.

---

Combined Analysis (Deterministic + LLM)

Runs all scanners together with optional LLM validation.

How It Works

1. 14 source scanners run on AST (fast, works on incomplete code)
2. 10 IR scanners run on ThalIR (semantic, requires valid syntax)
3. 1 LLM scanner (optional) analyzes code for all vulnerability types
4. Correlation engine links related findings across scanners

Benefits

• Complete coverage - Different scanners catch different issues
• Higher confidence - Findings confirmed by multiple scanners get higher scores
• Novel patterns - LLM finds issues deterministic scanners miss
• Reproducible baseline - Deterministic scanners (14+10) produce consistent results

When to Use

Best for:

• Pre-deployment comprehensive audits
• High-value contracts requiring maximum confidence
• Second opinions on critical code
• Discovering both known and novel vulnerabilities

Works on incomplete code: Yes (source-level scanning is error-tolerant)

Example

export OPENAI_API_KEY="sk-..."
tameshi analyze MyContract.sol \
  --correlation-threshold 0.7 \
  --cross-validated-only

VSCode: Run Tameshi: Run All Scans from Command Palette

This shows only findings confirmed by multiple scanners (highest confidence).

---

Choosing the Right Mode

Decision Tree

Active development (incomplete code)? → Use source scanning

• Real-time feedback as you type (<1s)
• Works on syntactically invalid code
• On-save in VSCode (default)
• Pre-commit hooks

Need semantic precision? → Use IR scanning

• Complete, compilable code required
• Cross-function analysis
• Pre-deployment audits

Need maximum coverage? → Use combined mode

• Run all 25 scanners together
• High-value contracts
• Combine all three modes
• Cross-validate findings
• Compliance documentation

---

Scan Mode Configuration

CLI

Choose mode with flags:

# Source + IR (deterministic)
tameshi scan run -i MyContract.sol --suite deterministic

# All modes including LLM
tameshi analyze MyContract.sol

# LLM only
tameshi analyze MyContract.sol --llm-only

VSCode

Configure automatic scanning mode in settings:

{
  "tameshi.scan.onSave": "file",
  "tameshi.llm.enabled": true,
  "tameshi.llm.useIrScanning": false
}

Set useIrScanning: true to send ThalIR to LLMs instead of raw source.

---

Analysis Pipeline Details

Source Scanning Pipeline

1. Parse Solidity with tree-sitter
2. Build AST
3. Pattern match against vulnerability signatures
4. Generate findings

IR Scanning Pipeline

1. Parse Solidity
2. Transform to ThalIR
3. Build control flow graphs
4. Run data flow analysis
5. Pattern match on IR
6. Map findings back to source

LLM Scanning Pipeline

1. Extract code representation (source or IR)
2. Build security-focused prompt
3. Send to LLM API
4. Parse structured response
5. Generate findings with explanations

---

Next Steps

• ThalIR - Learn about the intermediate representation
• Scanners - See what each mode detects
• CLI Reference - Command options for each mode