marketing-automation-analysis

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes marketing automation data retrieved from external sources. Malicious instructions embedded in the data (such as email subjects or lead names) could potentially influence the agent's behavior during analysis.
  • Ingestion points: Data is ingested from BigQuery, Snowflake, or Databricks via warehouse CLI tools called by asa.sh.
  • Boundary markers: The agent is provided with clear behavioral rules to state facts and disclose interpretations of business terms.
  • Capability inventory: The skill can execute warehouse commands, write to a local profile directory, and generate interactive dashboards.
  • Sanitization: Data rendered in the visualization dashboard is processed using html.escape.
  • [DYNAMIC_EXECUTION]: The skill generates interactive dashboards by constructing a JavaScript-enabled HTML file at runtime.
  • Evidence: The generate-dashboard.py script takes a JSON payload that includes a script field. This JavaScript string is injected directly into a local HTML template to initialize Chart.js visualizations.
  • [EXTERNAL_DOWNLOADS]: The visualization dashboard fetches a common charting library from an external service.
  • Evidence: The HTML template in generate-dashboard.py includes a link to load Chart.js from https://cdn.jsdelivr.net/npm/chart.js@4/dist/chart.umd.min.js. This is standard behavior for web-based visualizations.
  • [COMMAND_EXECUTION]: The skill executes command-line interfaces for data warehouses and system utilities.
  • Evidence: asa.py uses subprocess.run to call tools like bq, snow, and databricks. It correctly utilizes list-based arguments to prevent command injection. It also uses platform-specific clipboard utilities (pbcopy, clip, xclip, xsel) to assist the user during the setup process.
  • [CREDENTIALS_UNSAFE]: The skill manages Fivetran API tokens and warehouse configuration. It follows security best practices by storing tokens in a restricted local file (~/.fivetran/skills/marketing-automation-analysis/credentials.json) with 0600 permissions. It also reads the host and profile names from ~/.databrickscfg to provide tailored authentication troubleshooting for Databricks users.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 09:23 PM