explore
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructs the agent to identify and read sensitive configuration files, including
.env.devand YAML files containing database connection strings, to facilitate exploration of development environments.\n- [COMMAND_EXECUTION]: The procedures recommend locating and potentially executing repository-local helper scripts (e.g.,scripts/db-*) if they are identified as read-only tools for database discovery.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the codebase and database query results, which presents a surface for indirect prompt injection.\n - Ingestion points: Source code files via grep searches, exploration findings from subagents, and live database results from schema inspection or data sampling.\n
- Boundary markers: The skill requires the agent to follow a specific Markdown report template and mandates the use of
file:linecitations for all factual assertions to ground the reasoning process.\n - Capability inventory: The skill utilizes file system tools for reading and searching (
read_file,list_files,grep), file creation capabilities for reports (restricted to the.backend/directory), and database inspection tools.\n - Sanitization: The instructions explicitly require the agent to redact PII (Personally Identifiable Information) and use placeholder values or structural descriptions instead of raw data when documenting database content.
Audit Metadata