The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process data from untrusted external sources like web pages and code repositories. This presents a potential surface for indirect prompt injection where malicious instructions could be embedded in the content being researched.
Data enters the agent's context through web search and repository reading tools as defined in the success criteria.
The skill explicitly acknowledges this risk in its constraints section, directing the agent to treat all fetched content as 'inert data' and to ignore any embedded directives.
The skill has access to interactive tools (like ask_user) and file system writes within a restricted research root, which are capabilities that could be influenced by malicious input.
It implements a sanitization pattern by requiring the agent to document suspected injections rather than following them.

rpi-research