claude-skills
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- COMMAND_EXECUTION (HIGH): The skill frontmatter
allowed-toolsincludesBash,Write, andEdit, bypassing user permission prompts for arbitrary shell command execution and file modification. - REMOTE_CODE_EXECUTION (HIGH): The 'Dynamic Context Injection' feature automatically executes shell commands contained within backticks (e.g.,
`git status`) when the skill is loaded, providing a direct path for RCE if the skill content is untrusted. - PROMPT_INJECTION (HIGH): The skill interpolates user input via the
$ARGUMENTSvariable without sanitization or boundary markers. Combined with unprompted high-privilege tool access, this creates a severe surface for indirect prompt injection attacks where malicious input can execute unauthorized commands.
Recommendations
- AI detected serious security threats
Audit Metadata