architect-and-verify
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements a strict 'read and test' boundary, explicitly informing the user that the agent will not make changes to any system (installations, firewall rules, account creation).
- [SAFE]: Instructions mandate the total exclusion of sensitive data, stating that the document and recorded results must never hold passwords, keys, or tokens.
- [SAFE]: Network operations are restricted to specific connections defined in the architecture table to avoid unauthorized scanning and the potential triggering of intrusion-detection systems.
- [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: User-supplied environment names, system descriptions, and connectivity facts in Phases 1, 3, and 4.
- Boundary markers: A mandatory boundary statement is used to define the agent's limited role.
- Capability inventory: The skill utilizes connectivity diagnostic tools (
nc,curl,Test-NetConnection), system management utilities (systemctl,Get-Service), and database client commands (sqlcmd,SELECTqueries). - Sanitization: While the skill instructs the agent to use plain language, there are no specific requirements for sanitizing or escaping user-provided strings before they are interpolated into diagnostic shell commands.
Audit Metadata