external-model-consult
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a robust security model for delegated model consultation. It includes explicit instructions to sanitize inputs by removing secrets (API keys, tokens, private keys) and repository context before transmission to external providers.
- [COMMAND_EXECUTION]: The skill facilitates interaction with external providers through local CLI tools (OpenAI Codex and Google Antigravity). These operations are restricted to isolated, temporary working directories with the narrowest possible permissions to prevent unauthorized file system access.
- [EXTERNAL_DOWNLOADS]: The skill interfaces with well-known, trusted AI providers. The instructions prioritize user-managed authentication and explicitly prohibit the agent from bypassing permission filters or managing paid quotas without direct authorization.
- [SAFE]: Acknowledging the risk of indirect prompt injection, the instructions mandate that all data returned from external models be treated as untrusted. The agent is explicitly forbidden from auto-executing instructions or applying changes from these external sources without independent verification.
Audit Metadata