provider-onboarding
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill defines a 'Delegated Credential Refresh' mechanism that involves executing external vendor CLI tools, such as
claude doctor,grok models, andapp-server. It specifies safety constraints like process group detachment, non-interactive execution, and time budgets for these operations. - [INDIRECT_PROMPT_INJECTION]: The skill describes processes for ingesting and handling data from external providers (e.g., via HTTP responses and CLI outputs), which constitutes a vulnerability surface if those sources are compromised.
- Ingestion points: Data enters the system through
providerFetch(HTTP requests) and the outputs of delegated CLI commands (SKILL.md). - Boundary markers: The documentation references 'typed local resolution boundaries' and specific field tolerance rules for parsing provider responses (SKILL.md).
- Capability inventory: The system possesses network access, process table inspection capabilities (
ps), and the ability to execute specific vendor CLI commands (SKILL.md). - Sanitization: The skill mandates that refresh tokens must never be logged, cached, or rendered, and that output must be redacted to ensure secrets are not exposed (SKILL.md).
Audit Metadata