herdr-runtime
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the
herdrCLI to manage workspaces and agent sessions via commands likeherdr workspace createandherdr agent start. - [EXTERNAL_DOWNLOADS]: The skill instructs the agent to install integrations using
herdr integration install <kind>, which involves fetching and adding new capabilities to the environment. - [INDIRECT_PROMPT_INJECTION]: The skill functions as a supervisor, reading outputs from worker agents to coordinate tasks, which is a vector for indirect instructions. \n
- Ingestion points: Worker output is read via
herdr agent read. \n - Boundary markers: The documentation suggests using JSON ID return values for resource management to identify resources correctly. \n
- Capability inventory: Spawning processes, directory creation, and network communication via orchestrated agents. \n
- Sanitization: Recommends using explicitly returned IDs from the management tool to avoid state confusion and resource hijacking.
Audit Metadata