omh-agent-board
Warn
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill implements a meta-orchestration pattern where the 'prepare' tool returns a 'native_action' which the agent is then directed to invoke. This dynamic execution of generated tool calls can be used to bypass static analysis of available tool chains.
- [DYNAMIC_EXECUTION]: The skill uses a 'skill_view' function to dynamically load content from 'references/skill-common-rail.md' at runtime, which constitutes dynamic loading from a local path.
- [COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands using the 'omh' CLI tool ('omh runtime record') to track status on the host system. While these are vendor-provided tools, the execution of shell commands increases the skill's capability tier.
- [INDIRECT_PROMPT_INJECTION]: The skill serves as a high-level orchestrator that processes user-supplied requests to delegate tasks to other agents, creating an attack surface where malicious instructions in a user's task description could influence the behavior of coordinated sub-agents. 1. Ingestion points: User prompts through the chat interface. 2. Boundary markers: Prose references to 'prepared-vs-observed' boundaries are present, but no formal structural delimiters are defined for data interpolation. 3. Capability inventory: Access to 'omh' CLI for status recording, board management (create, link, comment), and delegation to other profiles. 4. Sanitization: Relies on the 'prepare' tool's schema validation; no specific sanitization of user-provided strings is mentioned.
Audit Metadata