gap-analysis

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes multiple Python code blocks via the shell using a command path defined in shasta.config.json. This is used to initialize the database and run the compliance scanner.
  • [DATA_EXFILTRATION]: The skill accesses AWS credentials and infrastructure metadata to perform the SOC 2 scan. This behavior is aligned with the skill's stated purpose of compliance auditing.
  • [PROMPT_INJECTION]: The skill processes and presents data from AWS scan findings and generated reports. This creates an indirect prompt injection surface where malicious content (e.g., in AWS resource tags or names) could attempt to influence the agent's behavior.
  • Ingestion points: AWS scan data via shasta.db.schema and reports from shasta.reports.generator.
  • Boundary markers: No explicit delimiters or instructions to ignore embedded content are present when the agent reads the Markdown report.
  • Capability inventory: The agent has the ability to execute shell commands (python -c), read/write local files, and interact with the AWS API.
  • Sanitization: There is no evidence of sanitization for findings retrieved from the external AWS environment before they are presented to the consultant-persona agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 05:24 PM
Security Audit — agent-trust-hub — gap-analysis