Skip to content

SECURITY_AUDIT.md

Latest commit

 

History

History
308 lines (229 loc) · 11.1 KB

SECURITY_AUDIT.md

File metadata and controls

308 lines (229 loc) · 11.1 KB
Raw

Security Audit Report

Date: 2026-02-02 (Updated) Version: 0.3.3 Status: ✅ COMPREHENSIVE SECURITY HARDENING COMPLETE

Executive Summary

Category Critical High Medium Low Total Fixed
npm Dependencies 0 0 0 0 0 ✅ 34/34
Code Vulnerabilities 0 0 0 0 0 ✅ ALL FIXED
Total 0 0 0 0 0 ✅

Result: All security issues fixed. Production-ready with comprehensive hardening.

1. npm Dependency Vulnerabilities - ALL FIXED

Fixed via npm overrides in package.json:

Vulnerability Severity Original Package Fix Applied
axios CSRF/SSRF HIGH @orca-so/whirlpool-sdk Override to axios ^1.7.4
bigint-buffer overflow HIGH Solana packages Override to @vekexasia/bigint-buffer2 ^1.0.4
elliptic crypto risk HIGH secp256k1 Replaced with @noble/secp256k1 ^3.0.0
nanoid predictable MODERATE @drift-labs/sdk Override to nanoid ^3.3.8
nodemailer DoS MODERATE Direct dependency Updated to ^7.0.13
undici DoS MODERATE discord.js Override to undici ^6.23.0
@cosmjs/crypto HIGH @wormhole-foundation/sdk Override to ^0.38.1 (uses @noble/curves)

npm audit result:

found 0 vulnerabilities

2. Code Vulnerabilities

✅ FIXED - HIGH Risk

2.1 Command Injection - Multiple Files ✅ FIXED

  • Original: execSync() with string interpolation allowing shell injection
  • Fix: Replaced with execFileSync() with array arguments across all files:
    • src/nodes/index.ts - notifications, clipboard, say, open, commandExists
    • src/process/index.ts - commandExists
    • src/permissions/index.ts - resolveCommandPath
    • src/hooks/index.ts - checkRequirements
    • src/daemon/index.ts - launchctl commands
    • src/macos/index.ts - runAppleScriptSync
    • src/agents/index.ts - exec_python
  • Status: ALL FIXED - 15+ injection points remediated

2.2 Unsafe Sandbox - src/security/index.ts ✅ FIXED

  • Original: new Function() sandbox is bypassable
  • Fix: Sandbox now DISABLED by default. Requires ALLOW_UNSAFE_SANDBOX=true env var to enable.
  • Status: ✅ FIXED - Secure by default

2.3 Weak Random ID Generation ✅ FIXED (Feb 2026)

  • Original: Math.random().toString(36) used for IDs across 21+ files
  • Fix: Created src/utils/id.ts with crypto.randomBytes() based generators
  • Files Fixed: alerts, usage, memory, media, cron, hooks, arbitrage, canvas, embeddings, agents, all extensions
  • Status: ✅ FIXED - All IDs now cryptographically secure

2.4 Canvas eval() Remote Code Execution ✅ FIXED (Feb 2026)

  • Original: Browser-side eval() executed arbitrary JS from WebSocket
  • Fix: JS eval disabled by default. Requires CANVAS_ALLOW_JS_EVAL=true to enable.
  • Status: ✅ FIXED - Secure by default

2.5 Task Runner Shell Injection ✅ FIXED (Feb 2026)

  • Original: spawn() with shell: true allowed command injection
  • Fix: Replaced with execFile(), added command validation, restricted env vars
  • Status: ✅ FIXED - No shell interpretation

2.6 XSS in Canvas Components ✅ FIXED (Feb 2026)

  • Original: User input rendered via innerHTML without sanitization
  • Fix: Added escapeHtml() and sanitizeStyle() to all component renderers
  • Status: ✅ FIXED - All output escaped

2.7 CORS Misconfiguration ✅ FIXED (Feb 2026)

  • Original: Access-Control-Allow-Credentials: true sent with wildcard origin
  • Fix: Credentials only allowed with specific origin allowlist
  • Status: ✅ FIXED - Proper CORS handling

2.8 Missing Rate Limiting ✅ FIXED (Feb 2026)

  • Original: No IP-based rate limiting on gateway
  • Fix: Added sliding window rate limiter (100 req/min default, configurable)
  • Status: ✅ FIXED - CLODDS_IP_RATE_LIMIT env var

2.9 Missing Security Headers ✅ FIXED (Feb 2026)

  • Original: No HSTS, X-Frame-Options, etc.
  • Fix: Added all recommended security headers + HTTPS enforcement option
  • Status: ✅ FIXED - CLODDS_HSTS_ENABLED, CLODDS_FORCE_HTTPS env vars

2.10 WebSocket Message Validation ✅ FIXED (Feb 2026)

  • Original: No validation of incoming WebSocket message structure
  • Fix: Added isValidWebMessage() validator + 1MB size limit
  • Status: ✅ FIXED - Type + size validation

2.11 Weak Session IDs ✅ FIXED (Feb 2026)

  • Original: Session IDs used Math.random()
  • Fix: Now uses crypto.randomBytes(16).toString('hex')
  • Status: ✅ FIXED - Cryptographically secure sessions

2.12 Escrow Keypairs In-Memory Only ✅ FIXED (Feb 2026)

  • Original: Escrow keypairs stored only in memory (Map<string, Keypair>)
  • Risk: Server restart = lost keypairs = funds unrecoverable
  • Fix: Keypairs now encrypted (AES-256-GCM) and stored in SQLite database
  • Encryption: Uses CLODDS_ESCROW_KEY or CLODDS_CREDENTIAL_KEY env var
  • Status: ✅ FIXED - Keypairs survive server restarts

All Previously Accepted Risks - NOW FIXED

Risk Original Status New Status
Rate limiting Recommended ✅ IMPLEMENTED
Math.random() IDs Accepted ✅ FIXED
Input validation Accepted ✅ IMPROVED
Security headers Recommended ✅ IMPLEMENTED

3. Remediation Summary

All Issues - FIXED

# Issue Fix Status
1 nodemailer Updated to 7.0.13 ✅ DONE
2 Command injection execFileSync with array args ✅ DONE
3 Unsafe sandbox Disabled by default ✅ DONE
4 elliptic/secp256k1 Replaced with @noble/secp256k1 ✅ DONE
5 bigint-buffer Override to @vekexasia/bigint-buffer2 ✅ DONE
6 axios in orca-sdk Override to axios ^1.7.4 ✅ DONE
7 discord.js undici Override to undici ^6.23.0 ✅ DONE
8 nanoid Override to nanoid ^3.3.8 ✅ DONE
9 @cosmjs/* elliptic Override to ^0.38.1 ✅ DONE
10 Math.random() IDs crypto.randomBytes() in 21 files ✅ DONE
11 Canvas eval() Disabled by default ✅ DONE
12 Task runner shell injection execFile + validation ✅ DONE
13 XSS in canvas HTML escaping ✅ DONE
14 CORS misconfiguration Proper credential handling ✅ DONE
15 Missing rate limiting IP-based sliding window ✅ DONE
16 Missing security headers HSTS, X-Frame-Options, etc. ✅ DONE
17 WebSocket validation Type + size validation ✅ DONE
18 Weak session IDs crypto.randomBytes() ✅ DONE

No Remaining Issues

All previously identified issues have been remediated.

4. Security Best Practices Implemented

✅ Encrypted credentials - AES-256-GCM at rest ✅ No hardcoded secrets - All from environment ✅ HTTPS enforced - For all API calls ✅ Webhook signature verification - HMAC validation ✅ SQL injection prevention - Parameterized queries ✅ Audit logging - All trades logged ✅ Modern crypto libraries - @noble/* instead of deprecated elliptic ✅ Security Shield - Code scanner (75 rules/9 categories), scam DB (70+ addresses from Etherscan/Mandiant/CertiK), multi-chain address checker, pre-trade tx validator, input sanitizer (zero-width/RTL/homoglyph/prompt injection detection)

5. Security Headers - IMPLEMENTED

The following headers are now automatically added by the gateway:

// Always added:
'X-Content-Type-Options': 'nosniff'
'X-Frame-Options': 'DENY'
'X-XSS-Protection': '1; mode=block'

// When CLODDS_HSTS_ENABLED=true or connection is HTTPS:
'Strict-Transport-Security': 'max-age=31536000; includeSubDomains'

Configuration

# Enable HSTS header
CLODDS_HSTS_ENABLED=true

# Force HTTP to HTTPS redirect
CLODDS_FORCE_HTTPS=true

# IP rate limiting (requests per minute)
CLODDS_IP_RATE_LIMIT=100

MCP Server Security

The MCP stdio server includes 5 security layers, all opt-in via environment variables:

Layer Env Var Default Description
Tool allowlist CLODDS_MCP_ALLOWED_TOOLS (all) Comma-separated list of allowed tool names
Tool blocklist CLODDS_MCP_BLOCKED_TOOLS (none) Comma-separated list of blocked tool names
Tool profiles CLODDS_MCP_TOOL_PROFILE full Predefined access: read-only, trading, full
Rate limiting CLODDS_MCP_RATE_LIMIT 60 Max calls per minute per client
Audit logging CLODDS_MCP_AUDIT true Structured JSON audit logs to stderr

Input sanitization runs automatically on all tool calls — string arguments are scanned for SQL injection, command injection, XSS, and path traversal patterns using the existing detectInjection() function.

# Example: restrict MCP to read-only tools with audit logging
CLODDS_MCP_TOOL_PROFILE=read-only
CLODDS_MCP_RATE_LIMIT=30
CLODDS_MCP_AUDIT=true

6. Publishing Checklist ✅

  • Fix nodemailer vulnerability
  • Fix command injection in nodes/index.ts
  • Add sandbox warning
  • Fix all npm audit vulnerabilities (34 → 0)
  • Replace elliptic with @noble/secp256k1
  • Override bigint-buffer with secure fork
  • Test all trading functions work after updates
  • Run npm audit - shows 0 vulnerabilities

7. Disclosure Policy

Security issues should be reported to: security@clodds.dev (or GitHub Security Advisories)

Do NOT create public issues for security vulnerabilities.

Security audit completed on 2026-01-30 All 34 npm vulnerabilities fixed Ready for npm publish

8. Server Hardening CLI

Clodds includes a built-in server hardening command for production deployments.

Usage

# Apply all hardening with interactive prompts
clodds secure

# Preview changes without modifying
clodds secure --dry-run

# Run security audit only
clodds secure audit

# Non-interactive mode (skip prompts)
clodds secure --yes

# Custom SSH port
clodds secure --ssh-port=2222

# Skip specific components
clodds secure --skip-firewall --skip-fail2ban

What it hardens

Component Changes Applied
SSH Disable password auth, root login, MaxAuthTries=3
Firewall (ufw) Allow SSH + custom ports, deny incoming by default
fail2ban Protect against brute force (5 failures = 1hr ban)
Auto-updates Enable unattended-upgrades for security patches
Kernel sysctl hardening (SYN cookies, ICMP redirects, etc.)

Security Audit Output

$ clodds secure audit

🔒 Clodds Server Security Hardening

ℹ === Security Audit ===

✔ SSH Password Auth: Disabled
✔ SSH Root Login: Disabled
⚠ SSH Port: Port 22 (consider changing from default)
✔ Firewall (ufw): Active
✔ fail2ban: Active (1 jail)
✔ Auto-updates: Configured

5 passed, 1 warnings, 0 failed

Post-Hardening Checklist

  1. Test SSH access in a new terminal before closing current session
  2. Verify firewall rules don't block required ports
  3. Monitor fail2ban logs for legitimate users being blocked
  4. Keep SSH key backed up securely