The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest, analyze, and summarize data from untrusted external sources such as PTT, Dcard, and various news and social media channels. This monitoring workflow creates a surface for indirect prompt injection, where attackers can embed instructions in public posts that the agent might interpret as directives when processing brand mentions or sentiment.
Ingestion points: The skill explicitly instructs the agent to fetch data via PTT search scripts, Dcard APIs, and news monitoring tools (documented in examples/sample_scenario.md and references/taiwan-media.md).
Boundary markers: The provided report templates and output formats lack explicit delimiters or instructions to the model to ignore any command-like text found within the monitored snippets.
Capability inventory: The defined workflow includes automated triggers that result in actions, such as sending LINE notifications to management or activating crisis protocols, which could be exploited by specific injection payloads.
Sanitization: The instructions do not specify any sanitization, filtering, or validation steps for the external content before it is processed by the model.

pr-media-monitoring