social-media

Indirect Prompt Injection

Potential vulnerability to indirect prompt injection via research data.

• Evidence: The skill mandates the use of a researcher sub-agent to gather information on a topic before drafting content. The instructions state: "Use the task tool with subagent_type: 'researcher'" and save the output to research/<topic>.md.
• Analysis: This creates an attack surface where a research sub-agent could unknowingly ingest malicious instructions from the web. These instructions could attempt to influence the subsequent drafting process performed by this skill. This pattern, known as tool output poisoning, is a common consideration for agents that integrate live research. The skill's capabilities are focused on file writing and image generation, and the risk is mitigated by standard agent guardrails.