architect
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python script located at '/scripts/roles.py' to manage runner roles during the design phase. The command uses a quoted shell heredoc to pass tool output securely, minimizing shell injection risks.
- [INDIRECT_PROMPT_INJECTION]: The skill utilizes a multi-agent workflow where a main agent synthesizes design packages from multiple child agents (runners). This creates an attack surface for indirect prompt injection.
- Ingestion points: Phase B (Sketch) ingests and synthesizes design packages from multiple 'arena' runners.
- Boundary markers: Runners are provided with specific instructions via 'references/runner-prompt.md' and their actions are logged and verifiable.
- Capability inventory: The skill possesses command execution capabilities (via python3) and file-writing capabilities during the implementation phase.
- Sanitization: A formal synthesis step and a design red-flag screening process ('references/design-red-flags.md') are used to validate and filter child agent outputs.
Audit Metadata