arcanea-coding-agent

Fail

Audited by Gen Agent Trust Hub on Feb 14, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONNO_CODE
Full Analysis
  • PROMPT_INJECTION (HIGH): The skill design exhibits a high-risk Indirect Prompt Injection surface (Category 8). \n
  • Ingestion points: Documentation specifies agents ingest untrusted data from the user's 'current file/project' and 'selected code' across any file. \n
  • Boundary markers: No delimiters or isolation techniques are described to separate untrusted content from system instructions. \n
  • Capability inventory: The agent can modify code ('improve this selected code') and execute 'opencode editor commands' (e.g., 'arcanea:invoke'). \n
  • Sanitization: No mention of sanitization or filtering for ingested file content. \n- COMMAND_EXECUTION (MEDIUM): The documentation defines several commands that integrate with the editor environment. Without implementation code, the safety of command argument parsing cannot be verified. \n- NO_CODE (INFO): The provided skill consists only of a markdown documentation file; all functional code, architecture, and UI files are missing, preventing verification of the stated capabilities.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 14, 2026, 05:52 PM