omh-systematic-review
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill orchestrates the 'omh' command-line utility to manage literature screening phases, specifically calling sub-commands like 'freeze', 'prepare', and 'merge' to handle local project files.
- [INDIRECT_PROMPT_INJECTION]: The workflow processes untrusted external content (research paper abstracts) which could potentially contain malicious instructions.
- Ingestion points: Untrusted content is ingested via the records JSONL file processed by the 'prepare' command.
- Boundary markers: No specific delimiters or 'ignore instructions' markers are mentioned for wrapping the abstract content.
- Capability inventory: The skill can execute local CLI commands and delegate tasks to subagents ('delegate_task').
- Sanitization: No explicit sanitization or filtering is defined for the records, although the dual-blind review process and required human verification provide significant mitigation against unintentional instruction following.
Audit Metadata