lov-init-auth
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes Python scripts (
profile_store.pyandvalidate_skill.py) to manage local state and validate source integrity. These scripts operate locally and include specific logic to handle atomic file writes and permission management (0o600 for config files).\n- [INDIRECT_PROMPT_INJECTION]: The skill acts as a code generator that processes application source code and configuration data.\n - Ingestion points: The skill reads React application source code, Supabase client configuration files, and local user profile data (
user-profile.md).\n - Boundary markers: The
SKILL.mdfile defines explicit execution boundaries that require the agent to use host-provided tools and parse requests within a clearly defined scope.\n - Capability inventory: The skill performs file system read/write operations for React project integration and executes its own internal Python scripts for lifecycle management.\n
- Sanitization: The
profile_store.pyscript implements a sanitization check using a denylist of sensitive keywords (e.g.,token,secret,password,api_key) to ensure that credentials are not persisted in the agent's long-term memory records.
Audit Metadata