molten-search
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the internet via tools like
WebSearch,WebFetch, and specialized research skills (e.g.,last30days,agent-reach).\n - Ingestion points: External content from websites, social media platforms (Reddit, X, TikTok, etc.), and code repositories are brought into the agent's context during the search and research process.\n
- Boundary markers: The instructions do not define clear delimiters or 'ignore' instructions for the retrieved data, potentially allowing instructions embedded in web content to influence the agent's behavior.\n
- Capability inventory: The skill has the ability to perform web searches, fetch URLs, read local skill files, and trigger the execution of other skill modules across multiple system directories.\n
- Sanitization: There is no explicit logic provided for sanitizing or filtering the content retrieved from external sources before it is processed by the agent.\n- [DYNAMIC_EXECUTION]: The skill implements dynamic instruction loading by resolving the paths of other skill modules at runtime. Based on the user's query, it searches through a predefined list of local and hidden directories (e.g.,
~/.agents/skills/,~/.cursor/skills/) to find and follow the instructions contained within a targetSKILL.mdfile.
Audit Metadata