The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to execute an embedded Python script block using the python3 -c command via the Bash tool to query Wikipedia and Wikidata. Running dynamically generated code at runtime is a risk factor as it can bypass static analysis and be manipulated to perform unintended actions.
[COMMAND_EXECUTION]: The Python script template provided in references/details.md interpolates a [Brand_Name] placeholder directly into a triple-quoted Python string. Because the instructions do not mandate sanitization of this input, a malicious brand name containing quotes, semicolons, or newlines could be used to break out of the string context and execute arbitrary Python code within the shell environment.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface by ingesting untrusted data from external sources (e.g., website robots.txt, llms.txt, and third-party platform search results) and using it alongside powerful tools like Bash and WebFetch.
Ingestion points: External site audit data (robots.txt, llms.txt) and search results from YouTube, Reddit, and LinkedIn.
Boundary markers: The instructions lack explicit delimiters or "ignore embedded instructions" warnings for the processed data.
Capability inventory: Uses Bash (shell access), Write (file system modification), and WebFetch (network operations).
Sanitization: There are no instructions for escaping or validating the content retrieved from external sites before processing or output generation.

geo-brand-mentions