How Trust Scores Work
Every tool in Forge is analyzed by Sigil's 8-phase security engine. Trust scores help you understand the security risk of AI agent tools before installation.
Trust Score Ranges
8-Phase Security Analysis
Install Hooks
Scans setup.py cmdclass, npm postinstall scripts, and Makefile targets for malicious install-time behavior
Code Patterns
Detects dangerous patterns like eval(), exec(), pickle.loads(), and child_process executions
Network / Exfil
Identifies outbound HTTP requests, webhooks, socket connections, and DNS tunneling attempts
Credentials
Searches for environment variable access, credential files (.aws, .kube), SSH keys, and API key patterns
Obfuscation
Detects Base64 encoding, character code manipulation, hex encoding, and minified suspicious payloads
Provenance
Analyzes Git history, author patterns, binary file inclusion, and hidden file presence
Prompt Injection
Detects jailbreak attempts, markdown-based RCE, and social engineering patterns
Skill Security
Scans for AI skill malware, skill.yaml tampering, and tool abuse patterns
Example Security Analysis
postgres-connector
HIGH TRUST(92)Clean database connector with standard SQL operations. No install hooks or suspicious patterns detected.
web-scraper-pro
MEDIUM TRUST(73)Makes external HTTP requests and processes dynamic content. Uses eval() for data parsing.
system-manager
LOW TRUST(28)Accesses system files, modifies PATH variables, and includes obfuscated code sections.
Analyze Your Tools with Sigil
Get detailed security reports, track tool updates, and collaborate with your team using Sigil's complete platform.