The Agent Skills Directory

[Unverifiable Dependencies] (MEDIUM): The requirements.txt file specifies package versions that do not match current official releases.
Evidence: typer==0.20.0 and xai-sdk==1.3.1 are specified. As of the current date, the latest stable version of typer is 0.12.x, and the official xai-sdk versions are significantly lower. Providing future-dated or non-existent version numbers is a common indicator of potential dependency confusion attacks.
[Indirect Prompt Injection] (LOW): The skill is designed to ingest and process untrusted data from the internet, creating a vulnerability to indirect prompt injection.
Ingestion points: The skill utilizes x_search() (Twitter) and web_search() to bring external, attacker-controllable content into the model's context.
Boundary markers: Absent. The script appends user queries and tool results without using delimiters or system-level instructions to ignore embedded commands in the retrieved data.
Capability inventory: The skill includes a code_execution tool (server-side Python) and web_search capabilities, which could be misused if the model is manipulated by injected instructions.
Sanitization: No sanitization, filtering, or validation is performed on the data returned from X/Twitter or the web before processing.
[Data Exposure & Exfiltration] (LOW): The skill transmits data to a third-party AI provider (xAI).
Evidence: User queries and tool results are sent to xAI servers. While necessary for the skill's function, this involves a non-whitelisted external domain and the transmission of potentially sensitive context to a third party.

xai-grok