personas-from-artifacts
Pass
Audited by Gen Agent Trust Hub on Jun 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes several commands through the coval CLI, including 'coval personas list', 'coval agents get', and 'coval personas create'. These commands are used to interact with the vendor's platform to manage agent configurations and persona data.
- [EXTERNAL_DOWNLOADS]: The skill's documentation references the coval CLI tool, an external dependency provided by the vendor and typically installed via system package managers like Homebrew.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes untrusted external artifacts (backend payloads, UI screenshots, product docs, and real user messages) to generate behavioral prompts for simulation personas.
- Ingestion points: Phase 2 in 'SKILL.md' describes the ingestion of various product artifacts to inform persona generation.
- Boundary markers: There are no explicit technical delimiters or instructions provided to the agent to ignore embedded commands within the processed artifacts.
- Capability inventory: The skill uses the coval CLI ('coval personas create' and 'coval personas update') to save the derived behavioral instructions.
- Sanitization: The skill includes a manual instruction for the user to "Strip PII before sharing" artifacts, but lacks automated sanitization for the content of the messages or payloads.
Audit Metadata