kanban-worker

Pass

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill instructs agents to read and adapt their behavior based on 'summaries', 'errors', and 'comment threads' from previous runs retrieved via the kanban_show tool. This represents an indirect prompt injection surface where a malicious actor or a compromised prior agent could inject instructions into these fields to influence the current agent's execution.
  • Ingestion points: Metadata and comments retrieved from the Kanban system via kanban_show (e.g., the runs list and comment history).
  • Boundary markers: The instructions do not define boundary markers or clear separations to distinguish between system guidance and potentially untrusted task data.
  • Capability inventory: The agent has the ability to modify files within its workspace ($HERMES_KANBAN_WORKSPACE), create new tasks via kanban_create, and mark tasks as complete or blocked with custom summaries.
  • Sanitization: There are no instructions for validating, sanitizing, or escaping the content retrieved from the Kanban system before it is processed or acted upon.
Audit Metadata
Risk Level
SAFE
Analyzed
May 16, 2026, 01:44 PM
Security Audit — agent-trust-hub — kanban-worker