architect
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes 'grounding artifacts' and task-specific data to synthesize architectures and eventually implement code. This creates a surface where instructions embedded in the analyzed codebases or grounding data could influence the agent's output or sub-agent behavior.
- Ingestion points: Phase A grounding artifacts (mental models of existing subsystems) and design-sketch task inputs in SKILL.md.
- Boundary markers: No specific delimiters or 'ignore instructions' warnings are defined for the ingested system data in the provided reference templates.
- Capability inventory: Phase D involves direct code implementation and file writing; Phase B orchestrates sub-agents through the
arenaskill. - Sanitization: The workflow does not describe validation or sanitization steps for grounding data before it is passed to the synthesis and implementation stages.
Audit Metadata