opencode-doc

Fail

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The documentation (SKILL.md, intro.md) explicitly recommends a highly insecure installation method where a script is fetched from a remote URL (https://opencode.ai/install) and piped directly into the bash shell. This pattern allows for unverified code execution on the user's host system.
  • [COMMAND_EXECUTION]: The skill documents a 'dynamic context injection' feature (commands.md, tui.md) that uses the !command`` syntax to execute shell commands at runtime. This allows arbitrary system commands to be triggered and their output injected into the agent's context.
  • [COMMAND_EXECUTION]: The custom command system documents a template engine where user-supplied input ($ARGUMENTS, $1, etc.) is interpolated into prompts. When combined with the documented shell execution syntax (!), this presents a significant command injection surface.
  • [DATA_EXFILTRATION]: The documentation provides examples of how the agent can be configured to access sensitive local files (e.g., ~/.secrets/openai-key) and environment variables, which increases the risk of credential exposure if the agent's instructions are manipulated.
Recommendations
  • HIGH: Downloads and executes remote code from: https://opencode.ai/install - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Jul 2, 2026, 07:11 PM
Security Audit — agent-trust-hub — opencode-doc