stof
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill teaches the AI agent to interpret and execute Stof documents which can contain embedded logic, creating an ingestion surface for potentially untrusted code.\n
- Ingestion points: The skill documents the use of
importandStd::parseto load and execute logic from external files or strings (SKILL.md, Std.md).\n - Boundary markers: The documentation explicitly advocates for the use of the
untrustedprofile and thesealfunction for confinement when handling external sources (SKILL.md, Std.md).\n - Capability inventory: The language supports HTTP fetch (Http.md), file system read/write (fs.md), environment variable access (Std.md), and dynamic code parsing (Std.md).\n
- Sanitization: The guide provides clear instructions on restricting library access and using object-level sandboxing (confinement) to protect the host document (SKILL.md).\n- [EXTERNAL_DOWNLOADS]: The
Httplibrary provides the agent with functions to perform asynchronous network requests to external domains.\n - Evidence: Documentation of
Http::fetchinreferences/libs/Http.md.\n- [COMMAND_EXECUTION]: The skill documents capabilities for the agent to interact with the local file system and manage environment variables.\n - Evidence:
fs::read,fs::write, andfs::read_stringdocumented inreferences/libs/fs.md;Std::envandStd::set_envdocumented inreferences/libs/Std.md.\n- [DYNAMIC_EXECUTION]: The skill includes functions that allow for the parsing and execution of Stof code strings at runtime, which is a core feature of the runtime.\n - Evidence:
Std::parseandRuntime::call_sourceare documented as primary methods for dynamic logic processing (SKILL.md, Std.md).
Audit Metadata