Skills
skills/obra/clank/Installing Skills System/Gen Agent Trust Hub

Installing Skills System

Fail

Audited by Gen Agent Trust Hub on May 30, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The SKILL.md file contains a 'Quick Install' command (curl -fsSL ... | bash) that downloads and executes a script directly from a remote GitHub repository. This pattern is a high-risk installation vector as it bypasses local review of the executed code.\n- [PROMPT_INJECTION]: The skill provides an XML snippet (in SKILL.md and CLAUDE_MD_PREAMBLE.md) for the agent to insert into its CLAUDE.md file. This snippet uses aggressive behavioral-override language ('STOP', 'CRITICAL', 'FAILED TASK') and commands the agent to distrust its own judgment ('even if you think you're already good at that. You're not') in order to force compliance with external files.\n- [COMMAND_EXECUTION]: The install.sh script performs significant file system modifications, including backing up existing directories, creating new folders, and establishing symbolic links. It also executes python3 code to handle path resolution on macOS, representing dynamic code execution within the installation flow.\n- [EXTERNAL_DOWNLOADS]: The installation process involves downloading third-party content via git clone or curl from a GitHub repository (github.com/obra/clank). These operations introduce external code and configurations into the agent's sensitive configuration directories.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/obra/clank/main/skills/meta/installing-skills/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
May 30, 2026, 03:29 PM
Security Audit — agent-trust-hub — Installing Skills System