9-Point Security Scan
Every agent on AgentXchange undergoes automated security analysis before listing. Our scanner checks for the most dangerous attack vectors in AI agent code.
Agents Scanned
Avg Trust Score
Trusted (80+)
Scan Coverage
Scan Results Breakdown
How It Works
Agent Submitted
Source code uploaded
9 Security Checks
Pattern matching + analysis
Trust Score
0-100 safety rating
Trust Score Calculation
Every agent starts at 100. Findings deduct points based on severity. The final score determines the safety tier.
Trusted
No critical or high findings. Safe to install.
Caution
Some findings detected. Review before installing.
Warning
Critical findings detected. Not recommended.
The 9 Security Checks
SSRF Detection
criticalScans for Server-Side Request Forgery patterns including cloud metadata endpoint access (169.254.169.254), private IP range targeting, and unvalidated fetch calls that could expose internal infrastructure.
Detects
- Cloud metadata harvesting
- Private network scanning
- Internal API access
Prompt Injection
criticalDetects attempts to override system instructions, hijack AI roles, inject hidden directives, or use invisible Unicode tags (U+E0000-E007F) to manipulate agent behavior.
Detects
- Instruction overrides
- Role hijacking
- Secrecy directives
- Invisible Unicode tags
Data Exfiltration
criticalIdentifies patterns that steal sensitive data — SSH keys, environment files, browser credentials — and exfiltrate them via webhook services, DNS tunneling, or steganography.
Detects
- SSH key theft
- .env file access
- Webhook exfiltration (ngrok, webhook.site)
- Crypto wallet targeting
Dangerous Commands
criticalFlags destructive system commands including recursive deletion (rm -rf /), fork bombs, reverse shells, and curl-pipe-bash execution chains.
Detects
- rm -rf /
- Fork bombs
- Reverse shells
- curl | bash pipes
Secret Detection
highScans for hardcoded credentials and API keys — AWS access keys (AKIA pattern), GitHub PATs, OpenAI/Anthropic keys, Stripe keys, JWTs, and private key material.
Detects
- AWS keys (AKIA...)
- GitHub PATs (ghp_)
- API keys (OpenAI, Anthropic, Stripe)
- Private keys & JWTs
Obfuscation
highDetects code obfuscation techniques used to hide malicious payloads — base64-encoded commands, eval/exec calls, hex-encoded strings, and dynamic code generation.
Detects
- Base64 payloads (atob, Buffer.from)
- eval() / exec()
- Hex string obfuscation
External Fetches
mediumIdentifies dynamic HTTP fetching patterns that could download and execute arbitrary code at runtime, bypassing static analysis.
Detects
- Dynamic fetch/axios calls
- Runtime code loading
- Unvalidated URL construction
Credential Access
highDetects attempts to access system credentials — reading sensitive environment variables, browser credential stores, keychain access, and password file enumeration.
Detects
- Sensitive env vars (DATABASE_URL, etc.)
- Browser credential paths
- Keychain access
Privilege Escalation
criticalFlags attempts to gain elevated system access — sudo commands, LaunchAgent installation, systemd service creation, and SUID bit modification.
Detects
- sudo / su commands
- LaunchAgent persistence
- systemd service injection
- SUID modifications
Every agent scanned. Every publisher verified.
AgentXchange is the only marketplace that security-scans every listing before it goes live. Trust is not optional — it's built in.