verify

Fail

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructs the user to execute the agent using the --permission-mode bypassPermissions flag. This configuration explicitly removes the human-in-the-loop requirement, granting the agent the ability to perform sensitive actions—such as running shell commands or modifying the local filesystem—without user confirmation.
  • [COMMAND_EXECUTION]: The provided shell command snippet facilitates the execution of a powerful CLI tool (claude) with full system access capabilities. By bypassing permission gates, the skill creates an environment where the agent can execute any tool it deems necessary, essentially allowing autonomous code execution on the host machine.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data, specifically the user's prompt, which is interpolated directly into the agent's command line.
  • Ingestion points: The user's prompt is ingested as a command-line argument in the script provided in SKILL.md.
  • Boundary markers: The prompt is enclosed in quotes, which provides minimal protection against command injection and no protection against instructions embedded within the prompt itself.
  • Capability inventory: The claude CLI, when running with the provided configuration, has the capability to execute shell commands, read/write files, and perform network operations.
  • Sanitization: No sanitization or input validation logic is present to prevent the agent from obeying malicious instructions embedded in the user prompt while in 'bypass' mode.
  • [EXTERNAL_DOWNLOADS]: The documentation suggests installing the tiktoken library, which is a standard package from a well-known vendor used for token counting.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 21, 2026, 03:03 PM