skills/hainrixz/claude-db/introspect/Gen Agent Trust Hub

introspect

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to interact with database engines and fetch catalog information. This is consistent with its stated purpose of live database introspection.
  • [DATA_EXFILTRATION]: The skill accesses the $DATABASE_URL environment variable for database connectivity. It includes instructions to redact credentials from logs and findings using a redactSecrets() function to prevent accidental exposure.
  • [PROMPT_INJECTION]: The skill processes untrusted metadata from external databases, creating an indirect prompt injection surface. Ingestion points: Database catalogs (Postgres, MySQL, Mongo, Cassandra) fetched via Bash or MCP tools. Boundary markers: Not explicitly specified for the ingested metadata content. Capability inventory: Bash (shell access) and Read (file access) tools. Sanitization: Includes credential redaction but lacks specific sanitization for data retrieved from catalog queries.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 10:01 PM
Security Audit — agent-trust-hub — introspect