audit-db-schema
Warn
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructions direct the agent to read sensitive environment files (
.envand.env.local) to extract the Supabase Project ID from theSUPABASE_URL. While this is used to correctly scope database connection parameters for auditing, environment files typically contain highly sensitive credentials and secrets. - [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from both database metadata and external web sources, creating a vulnerability to indirect prompt injection.
- Ingestion points: Database schema metadata (table names, column definitions, RLS policies) retrieved via the
supabase:execute_sqltool, and web content scraped from search results usingfirecrawl:firecrawl_scrape. - Boundary markers: The instructions do not mandate the use of delimiters, XML-like tags, or specific 'ignore embedded instructions' warnings when handling data retrieved from the database or the web.
- Capability inventory: The agent possesses the capability to execute arbitrary SQL queries and perform subsequent web searches, which could be abused via a multi-step injection attack.
- Sanitization: There is no mention of sanitizing, validating, or escaping the external data before it is interpolated into the final audit report or processed by the model.
Audit Metadata