Skills
Navigate
skills/aiagentskills/skills/model-debugging/Gen Agent Trust Hub

model-debugging

Fail

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The SKILL.md file contains a hardcoded Tinybird API token (p.eyJ1I...) intended for querying model health pipes.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute privileged administrative commands on remote servers via SSH, including sudo journalctl for log inspection and sudo systemctl restart for service management.
  • [DATA_EXFILTRATION]: Instructions are provided to access and read sensitive .env files and SOPS-encrypted secrets containing Azure, Google Cloud, and Cloudflare API keys and account identifiers.
  • [COMMAND_EXECUTION]: The shell scripts located in the scripts/ directory (e.g., check-user-errors.sh, find-403-users.sh) are vulnerable to SQL injection. They interpolate the $USERNAME and $TIER_FILTER shell arguments directly into SQL query strings sent to the Tinybird API without sanitization.
  • [COMMAND_EXECUTION]: The skill uses wrangler d1 execute to run arbitrary SQL commands against a production database (pollinations-db).
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 6, 2026, 04:52 PM
Security Audit — agent-trust-hub — model-debugging