The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to aggressively scan and analyze untrusted external sources, including social media whispers and news leaks, which are potential vectors for hidden instructions intended to manipulate agent behavior.
Ingestion points: Financial news, social media platforms, and SEC filing data processed via the rumor_scanner.py script.
Boundary markers: The skill relies on basic Markdown formatting within the prompt to separate instructions from data, providing no strong isolation or explicit instructions to ignore embedded commands in the source material.
Capability inventory: The SYSTEM_PROMPT explicitly directs the agent to "use your tools" to fetch and search live data, potentially enabling the execution of malicious instructions found in analyzed content if the agent has active tool capabilities (e.g., file access or network operations).
Sanitization: There is no evidence of input sanitization or filtering applied to the external content before it is processed by the model.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the openai Python package and communicates with the api.aisa.one endpoint. These are functional requirements for the skill's operation and are associated with the vendor's infrastructure.

stock-rumors