agent-architect
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides a framework for agents that process untrusted external data, establishing a potential attack surface.
- Ingestion points: The design workflow specifies ingestion of user task inputs, external reference documents, and dynamic tool outputs (references/context-engineering.md).
- Boundary markers: The skill recommends architectural constraints and middleware for enforcement, though specific prompt-level delimiters are not defined in the provided templates.
- Capability inventory: The underlying framework supports significant capabilities including filesystem access, shell execution, and remote service interop via MCP and REST (references/ol-ai-services-map.md).
- Sanitization: The guidelines suggest validation layers and middleware enforcement but rely on the implementer to provide specific data sanitization logic.
- [COMMAND_EXECUTION]: The architecture map documents a built-in shell tool for system-level command execution, intended as a capability for designed agents within the ol_ai_services ecosystem.
- [SAFE]: All external URL references target official documentation and repositories within the OntoLedgy organization's verified GitHub presence.
Audit Metadata