x07-language-guide

Warn

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The file references/language-guide.md contains imperative instructions intended to bypass standard AI safety and formatting protocols, such as: "Output ONLY one JSON object (no preamble)" and "Do NOT wrap the JSON in Markdown code fences."
  • [COMMAND_EXECUTION]: The language specification defines high-risk system-level APIs within the std.os.* and os.* namespaces, providing an attack surface for reading/writing files (os.fs.read_file, os.fs.write_file), environment variable access (os.env.get), and process execution (os.process.spawn_capture_v1).
  • [DATA_EXFILTRATION]: The documentation describes networking capabilities through os.net.http_request, which represents a potential risk for data exfiltration if the agent is used to process sensitive information via the X07 language.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 14, 2026, 03:09 PM