t3code
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill supports the execution of background Bash commands for managing development servers and long-running processes.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from external sources, such as task briefs and pull request comments, which creates a surface for indirect instructions.
- Ingestion points: Task briefs referenced in
delegate_task, pull request comments monitored viawatch_pull_request, and messages from peer threads. - Boundary markers: The skill suggests using external brief files rather than inline text, though specific delimiters for the content of those files are not defined.
- Capability inventory: Background Bash execution, file system access via git worktrees, and the ability to delegate tasks to other agent models.
- Sanitization: No specific input sanitization or filtering logic is defined for the content of processed briefs or comments.
- [PRIVILEGE_ESCALATION]: The skill implements specific security logic to prevent privilege escalation by ensuring that child agents and launched threads inherit restricted runtime and interaction modes from the parent orchestrator.
Audit Metadata