gre-reading-comprehension

Pass

Audited by Gen Agent Trust Hub on Jul 19, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill consists solely of markdown instructions and evaluation datasets. There are no scripts, binaries, or configuration files that could execute code.
  • [EXTERNAL_DOWNLOADS]: No external URLs, remote dependencies, or package managers (npm, pip) are used. All content is local to the skill package.
  • [DATA_EXFILTRATION]: There is no evidence of network activity, credential harvesting, or access to sensitive file paths (e.g., .ssh, .aws, .env).
  • [PROMPT_INJECTION]: The instructions do not contain any patterns attempting to bypass safety filters or override the agent's core behavioral constraints.
  • [COMMAND_EXECUTION]: No shell commands, subprocess calls, or system-level operations are present in the provided files.
  • [INDIRECT_PROMPT_INJECTION]: While the skill is designed to process user-supplied GRE passages and questions, it lacks the technical capabilities (such as file-system writes or internet access) that would allow an attacker to exploit the agent through embedded instructions in that data.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 19, 2026, 10:00 AM
Security Audit — agent-trust-hub — gre-reading-comprehension