swarm

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on the herdr CLI to perform its primary functions within a managed workspace.
  • Evidence: Use of commands such as herdr agent list, herdr agent prompt, herdr agent wait, and herdr agent read to manage agent tabs and communication.
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that forwards untrusted user input to other AI models, which can be exploited if the input contains malicious instructions.
  • Ingestion points: The user-supplied <task> parameter used in the herdr agent prompt command in SKILL.md.
  • Boundary markers: Absent. There are no explicit delimiters (e.g., XML tags or unique string markers) defined to separate the orchestrator's instructions from the user-provided task content.
  • Capability inventory: The skill has the ability to interact with the workspace, spawn new agent sessions, and read the output of various models.
  • Sanitization: The instructions do not include any steps for sanitizing, escaping, or validating the user input before it is relayed to the sub-agents.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 08:50 PM
Security Audit — agent-trust-hub — swarm