claude-code-dispatch
Warn
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
bashtool to invoke the Claude Code CLI tool (claude). - [PRIVILEGE_ESCALATION]: The instructions explicitly direct the agent to use the
--dangerously-skip-permissionsflag. This flag is designed to suppress all interactive confirmation prompts within the Claude Code environment. By using this flag, the agent bypasses the standard human-in-the-loop security controls, allowing the sub-agent to autonomously modify the local filesystem and execute terminal commands without oversight. - [INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by delegating complex tasks to a sub-agent using arbitrary prompt strings (
-p). If these strings incorporate untrusted data from the project or external sources, they could influence the sub-agent to perform unauthorized actions. The risk is elevated because the--dangerously-skip-permissionsflag removes the user's ability to review and block those actions. - Ingestion points: The prompt argument (
-p) for theclaudecommand inSKILL.md. - Boundary markers: Absent. The skill does not provide instructions to wrap the sub-task prompt in delimiters or warn the sub-agent about potentially malicious embedded instructions.
- Capability inventory: The sub-agent (Claude Code) has broad capabilities including file system read/write and arbitrary shell command execution.
- Sanitization: Absent. The skill does not describe any validation or escaping for the instructions passed to the sub-agent.
Audit Metadata