skills/ele-yufo/opencli/opencli-usage/Gen Agent Trust Hub

opencli-usage

Fail

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONCREDENTIALS_UNSAFEPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation includes commands for installing plugins from arbitrary GitHub repositories (opencli plugin install github:user/repo). This functionality allows the agent to fetch and execute third-party code from unverified external sources.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of software from non-trusted sources, specifically npm install -g @jackwener/opencli and git clone git@github.com:jackwener/OpenCLI.git. The author 'jackwener' is not a recognized trusted vendor.\n- [DYNAMIC_EXECUTION]: The skill describes a 'Self-repair' workflow where the agent is expected to patch local JavaScript adapter files (adapterSourcePath) and retry execution. This involves the agent dynamically modifying its own executable components. It also supports loading and executing custom JavaScript adapters stored in ~/.opencli/clis/.\n- [CREDENTIALS_UNSAFE]: The tool is designed to capture and utilize sensitive browser state, including cookies and intercepted authenticated requests (using strategies like COOKIE, INTERCEPT, and UI). This exposes highly sensitive authentication data to the agent and the underlying CLI tool.\n- [PERSISTENCE]: The opencli completion command generates shell scripts that are intended to be saved or sourced in the user's shell configuration (.bashrc, .zshrc, etc.), which can persistently alter the shell environment across sessions.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from arbitrary external websites and platforms (opencli <site> <command>), creating a surface for indirect prompt injection. \n
  • Ingestion points: Data is ingested through site adapters and browser extraction commands (opencli browser extract). \n
  • Boundary markers: None mentioned to distinguish between instructions and site content. \n
  • Capability inventory: The skill has shell access via Bash(opencli:*) and the ability to write to the file system to modify its own adapters. \n
  • Sanitization: No sanitization or filtering of external content is specified.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 16, 2026, 09:48 PM