specialist-engineering-software-architect
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Direct interpolation of user input into a subagent prompt. The skill uses the variable
{the user's question}at the end of theAgenttool's prompt argument inSKILL.md. This allows user-provided text to potentially override the architect persona's instructions. - [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through user-supplied questions. 1. Ingestion points: User input field
{the user's question}inSKILL.md. 2. Boundary markers: Absent; the user input is appended to the prompt without clear separators or isolation instructions. 3. Capability inventory: Spawns subagents using theAgenttool. 4. Sanitization: None; the input is passed directly to the subagent without filtering or validation.
Audit Metadata