Code Analysis

Field	Value
Document ID	ASCEND-SEC-004
Version	1.0.0
Last Updated	December 19, 2025
Author	Ascend Engineering Team
Publisher	OW-KAI Technologies Inc.
Classification	Enterprise Client Documentation
Compliance	SOC 2 CC6.1/CC6.2, PCI-DSS 7.1/8.3, HIPAA 164.312, NIST 800-53 AC-2/SI-4

Enterprise-grade code pattern detection for AI agent actions.

Overview

ASCEND provides real-time detection of dangerous code patterns in AI agent actions. The service analyzes SQL queries, shell commands, Python scripts, and JavaScript code for security vulnerabilities with compliance mappings to CWE, MITRE ATT&CK, and CVSS.

Architecture

Action Submission
       │
       ▼
┌─────────────────────────┐
│ CodeAnalysisService     │
├─────────────────────────┤
│ 1. Extract code         │  ← query, script, command params
│ 2. Detect language      │  ← SQL, Python, Shell, JS
│ 3. Load patterns        │  ← global + org custom
│ 4. Match patterns       │  ← regex analysis
│ 5. Calculate risk       │  ← severity_scores from config
│ 6. Block/allow          │  ← block_threshold from config
└─────────────────────────┘
       │
       ▼
   Governance Decision

Supported Languages

Language	Detection	Example Patterns
SQL	SELECT, INSERT, DROP, WHERE	SQL injection, destructive DDL
Python	def, class, import	eval/exec, subprocess, file access
Shell	echo, rm, cat, pipes	Command injection, privilege escalation
JavaScript	function, const, arrow functions	eval, DOM manipulation

Global Patterns (20 patterns)

ASCEND ships with 20 vendor-managed patterns stored in global_code_patterns. Categories:

Category	Description	Severity Range
injection	SQL injection, command injection	Critical-High
code_execution	Dynamic code execution (eval, exec)	Critical
code_injection	Script injection attempts	Critical
file_access	Unauthorized file operations	High-Medium
network_access	Outbound connections, data exfil	High
data_destruction	DROP TABLE, rm -rf, etc.	Critical
data_exfiltration	Data export, external transmission	High
credential_exposure	Password/key/token patterns	Critical-High
privilege_escalation	sudo, chmod, chown operations	High
sandbox_escape	Breakout attempts, dunder access	Critical

Key Patterns

Pattern ID	Language	Category	Severity	CVSS
SQL-001	sql	injection	critical	9.8
SQL-002	sql	data_destruction	critical	9.1
PY-001	python	code_execution	critical	9.8
PY-002	python	code_execution	critical	9.1
PY-004	python	code_injection	critical	9.5
SHELL-001	shell	code_injection	critical	9.8
SHELL-002	shell	data_destruction	critical	9.5

Configuration

All configuration is stored in org_code_analysis_config. No hardcoded values.

Default Settings

-- Default configuration created per-organization
enabled = true
mode = 'monitor'              -- 'enforce', 'monitor', 'off'
block_threshold = 90          -- Block if risk >= 90
alert_threshold = 50          -- Alert if risk >= 50
escalate_threshold = 70       -- Escalate if risk >= 70

-- Severity scores (configurable per-org)
severity_scores = {
  "critical": 95,
  "high": 75,
  "medium": 50,
  "low": 25,
  "info": 10
}

-- Language/category filters (empty = all enabled)
enabled_languages = []
enabled_categories = []
disabled_pattern_ids = []

Modes

Mode	Behavior
enforce	Detect AND block when critical findings exceed threshold
monitor	Detect and log, but never block
off	Disabled - no analysis performed

Pipeline Integration

Code analysis runs at Step 1.5 in the action submission pipeline:

POST /api/v1/actions/submit
├── Step 1:   Risk Enrichment
├── Step 1.5: CODE ANALYSIS ← Here
├── Step 1.6: Prompt Security (Phase 10)
├── Step 2:   CVSS Calculation
├── Step 3:   Policy Evaluation
└── ...

Usage

Direct Service Usage

from services.code_analysis_service import CodeAnalysisService

service = CodeAnalysisService(db, org_id=1)
result = service.analyze_for_action(
    action_type="execute_sql",
    parameters={"query": "SELECT * FROM users WHERE id = 1; DROP TABLE users;"},
    agent_id="agent-123"  # Optional: for agent-specific thresholds
)

if result.blocked:
    print(f"Blocked: {result.block_reason}")
else:
    print(f"Risk score: {result.max_risk_score}")
    for finding in result.findings:
        print(f"  - {finding.pattern_id}: {finding.description}")

Response Format

{
  "code_analysis": {
    "analyzed": true,
    "language": "sql",
    "findings_count": 2,
    "max_severity": "critical",
    "max_risk_score": 98,
    "patterns_matched": ["SQL-001", "SQL-002"],
    "blocked": true,
    "block_reason": "Critical code pattern detected: SQL-001 - SQL injection vulnerability",
    "config_mode": "enforce",
    "scan_duration_ms": 12,
    "findings": [
      {
        "pattern_id": "SQL-001",
        "severity": "critical",
        "category": "injection",
        "description": "SQL injection vulnerability",
        "matched_text": "'; DROP TABLE users;--",
        "line_number": 1,
        "cwe_ids": ["CWE-89"],
        "mitre_techniques": ["T1190"],
        "cvss_base_score": 9.8,
        "risk_score": 98
      }
    ]
  }
}

Code Parameter Detection

The service automatically extracts code from these parameter names:

Parameter	Description
query	SQL queries
sql	SQL statements
code	Generic code
script	Script content
command	Shell commands
cmd	Command aliases
statement	SQL/code statements
expression	Expressions to evaluate
source	Source code
shell	Shell scripts
bash	Bash commands
description	May contain code snippets

Custom Patterns

Organizations can add custom patterns with IDs prefixed CUSTOM-:

# Create custom pattern
pattern = OrgCustomPattern(
    organization_id=1,
    pattern_id="CUSTOM-PII-001",
    language="any",
    category="data_exfiltration",
    severity="high",
    pattern_type="regex",
    pattern_value=r"\b\d{3}-\d{2}-\d{4}\b",  # SSN pattern
    pattern_flags="IGNORECASE",
    description="Social Security Number detected",
    cwe_ids=["CWE-200"],
    mitre_techniques=["T1530"],
    cvss_base_score=7.5
)

Pattern Overrides

Disable or adjust global patterns for your organization:

# Create override
override = OrgPatternOverride(
    organization_id=1,
    pattern_id="SQL-003",
    is_disabled=False,
    severity_override="medium",
    risk_score_override=50,
    modified_by=user_id,
    modification_reason="False positives for our ORM-generated queries"
)

Agent-Specific Thresholds

Registered agents can have their own risk thresholds via RegisteredAgent.max_risk_threshold. The service uses the minimum of:

1. org_code_analysis_config.block_threshold
2. RegisteredAgent.max_risk_threshold

This allows restricting high-risk agents more than the org default.

Compliance

• CWE-78: OS Command Injection
• CWE-89: SQL Injection
• CWE-94: Code Injection
• CWE-95: Eval Injection
• CWE-200: Information Exposure
• MITRE ATT&CK: T1059, T1190, T1071, T1530
• NIST 800-53: SI-10 (Information Input Validation)
• SOC 2: CC6.1 (Access Control)

Troubleshooting

Code Not Being Analyzed

-- Check org config
SELECT enabled, mode
FROM org_code_analysis_config
WHERE organization_id = X;

Pattern Not Matching

-- Check if pattern is disabled
SELECT is_disabled
FROM org_pattern_overrides
WHERE organization_id = X AND pattern_id = 'SQL-001';

-- Check language filtering
SELECT enabled_languages
FROM org_code_analysis_config
WHERE organization_id = X;

Performance Issues

• Normal scan duration: less than 50ms
• If greater than 100ms, check pattern count and complexity
• Consider disabling unused languages/categories