skills/av/skills/run-llms/Gen Agent Trust Hub

run-llms

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [Remote Code Execution] (CRITICAL): Detected piped remote execution: 'curl https://av.codes/get-harbor.sh | bash' in SKILL.md. This allows an untrusted external source to execute arbitrary code with the user's permissions.
  • [Privilege Escalation] (HIGH): The skill requires 'sudo' for installation and instructs users to add themselves to the 'docker' group ('sudo usermod -aG docker $USER'), which is equivalent to granting root access.
  • [Indirect Prompt Injection] (HIGH): High risk of exploitation via untrusted inputs. Evidence: 1. Ingestion: 'harbor profile use ' and 'harbor pull' from external repositories. 2. Boundary markers: None. 3. Capability: Full container access via 'harbor shell' and 'harbor exec', plus host-level Docker access. 4. Sanitization: None detected.
  • [Persistence & Network Access] (MEDIUM): The 'harbor tunnel' command allows exposing local services to the internet, creating a potential callback mechanism. 'harbor profile use ' allows loading potentially malicious configurations from remote sources.
Recommendations
  • CRITICAL: Downloads and executes remote code from untrusted source(s): https://av.codes/get-harbor.sh - DO NOT USE
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Feb 16, 2026, 04:33 AM