check-agent-access

Warn

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill generates and executes administrative SQL GRANT statements to modify privileges within the Snowflake account (e.g., GRANT USAGE ON AGENT, GRANT SELECT ON TABLE). While this is the intended functionality, it provides a mechanism for privilege escalation if the agent's logic is manipulated or if it incorrectly identifies the necessary access levels.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes metadata and the instructions field from Cortex Agents to discover object dependencies. Malicious content within an agent's configuration could potentially influence the audit results or the remediation actions.
  • Ingestion points: Cortex Agent specifications and instructions fields retrieved via the semantic_studio tool (SKILL.md, Workflow Step 3).
  • Boundary markers: The skill lacks explicit instructions or delimiters to isolate untrusted agent metadata during processing.
  • Capability inventory: The skill possesses the capability to execute privilege-modifying commands through the snowflake_sql_execute tool.
  • Sanitization: No evidence is provided for sanitizing or validating the object names or dependency strings extracted from the agent specifications before they are used in SQL generation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 6, 2026, 03:06 AM
Security Audit — agent-trust-hub — check-agent-access