nexus
Warn
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The framework uses the Bash tool to coordinate sub-agents and external CLI tools. It specifically utilizes flags such as --dangerously-skip-permissions and bypassPermissions to enable autonomous loops that do not require human interaction, which can be highly risky if executed outside of isolated sandboxes.
- [INDIRECT_PROMPT_INJECTION]: Several recipes, notably newsroom and research, ingest external data from the web using WebFetch and WebSearch. This content is used for fact-checking and research grounding, and though the skill mentions verification steps, this ingestion represents a significant surface for indirect prompt injection from untrusted sources.
- [DYNAMIC_EXECUTION]: In the creative and podium workflows, the Builder agent is tasked with generating Python or JavaScript code for image generation. This code is intended to be executed in the publication phase to materialize final assets, constituting a dynamic code generation and execution pattern.
- [EXTERNAL_DOWNLOADS]: The skill performs broad research sweeps across the web to gather data from first-party documentation, technical articles, and other online sources to build its Evidence Ledger and verify claims.
Audit Metadata