architecture-alternatives

Pass

Audited by Gen Agent Trust Hub on Oct 22, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to provide the user with installation commands for external tools if they are not already available in the environment. Specifically, it points to resources from a third-party developer (Matt Pocock) via the commands: npx skills@latest add mattpocock/skills --skill grilling and npx skills@latest add mattpocock/skills --skill codebase-design. These commands are provided as text for the user to execute manually, rather than being executed silently by the agent.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect injection as it ingests untrusted requirements from the user during the 'grilling' phase and uses that data to generate code examples.\n
  • Ingestion points: User input provided during the 'Grill the requirements' phase in SKILL.md.\n
  • Boundary markers: Absent; there are no specific delimiters or warnings for the agent to ignore instructions embedded within the user's architectural requirements.\n
  • Capability inventory: The skill generates and displays code blocks ('Show the client code' section) based on user input, which could potentially be manipulated.\n
  • Sanitization: Absent; the skill does not specify filtering or escaping of the user-provided requirements before generating the architectural designs.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 22, 2026, 01:05 PM