skills/4ier/agent-skills/offload2/Gen Agent Trust Hub

offload2

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONNO_CODEPROMPT_INJECTION
Full Analysis
  • [NO_CODE] (HIGH): The core logic resides in scripts/offload.sh, which is not provided. This prevents verification of the script's safety and is a major security gap.
  • [COMMAND_EXECUTION] (HIGH): The skill executes a local shell script with arbitrary input. Without the source, it is impossible to determine if the script properly sanitizes input to prevent shell metacharacter injection.
  • [PROMPT_INJECTION] (HIGH): Category 8 Evidence: (1) Ingestion: Arbitrary text via stdin. (2) Boundary markers: None. (3) Capability: Local file write/append. (4) Sanitization: Unknown. The ingestion of untrusted reasoning state into the filesystem creates a significant indirect prompt injection surface.
  • [COMMAND_EXECUTION] (MEDIUM): The OFFLOAD2_FILE variable allows writing to arbitrary host paths. An attacker could potentially overwrite sensitive configuration files or achieve persistence by targeting shell profiles.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 01:16 PM