coval-eval-audit
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the
covalCLI (a tool provided by the vendor, coval-ai) to fetch configuration and results. The instructions specifically limit the agent to read-only operations such asget,list, andmanifestto prevent unintended modifications.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data, including conversation transcripts and metric explanations. While this represents a potential surface for indirect prompt injection, the risk is mitigated by the skill's read-only mandate and its specific focus on auditing rather than executing content from these sources.\n - Ingestion points: Simulated conversations, transcripts, and metric evidence fetched via the
covalCLI in SKILL.md.\n - Boundary markers: None explicitly defined for external data delimiters.\n
- Capability inventory: Restricted to read-only CLI commands (
coval --agent ... get/list).\n - Sanitization: The skill does not specify automated sanitization but instructs the agent to audit the data for specific decision risks.
Audit Metadata