design-persona
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute several
covalCLI commands, such ascoval whoami,coval personas list,coval agents list,coval agents get, andcoval personas create. These commands are used to manage testing assets on the vendor's platform and are consistent with the skill's stated purpose. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied information and agent configurations to generate CLI commands and behavior prompts, creating a surface for potential instruction injection.\n
- Ingestion points: External agent metadata via
coval agents get(SKILL.md, Phase 1) and user-provided persona descriptions in Phase 5.\n - Boundary markers: Not present; user inputs are interpolated directly into command arguments and prompt templates.\n
- Capability inventory: Includes shell command execution and potential network interaction via the
covalCLI.\n - Sanitization: No explicit sanitization steps are defined for external data before interpolation.
Audit Metadata