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

Installing Skills System

Fail

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: HIGHPROMPT_INJECTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill provides an XML block intended for the agent's persistent configuration file (~/.claude/CLAUDE.md) that uses forceful behavioral override language.
  • Evidence: SKILL.md and CLAUDE_MD_PREAMBLE.md contain instructions designed to bypass the agent's default task-handling logic: "STOP. Before doing ANYTHING else...", "CRITICAL: If a skill exists for your task, you MUST use it", and "SKILLS EXIST AND YOU DIDN'T USE THEM = FAILED TASK."
  • [REMOTE_CODE_EXECUTION]: The documentation includes a 'Quick Install' method for users that pipes a remote script directly to the bash shell.
  • Evidence: curl -fsSL https://raw.githubusercontent.com/obra/clank/main/skills/meta/installing-skills/install.sh | bash in SKILL.md. This script originates from the author's repository.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone a remote repository to the local file system.
  • Evidence: git clone https://github.com/obra/clank.git or gh repo fork obra/clank in SKILL.md.
  • [COMMAND_EXECUTION]: The install.sh script executes various shell commands to manipulate the local file system and configuration.
  • Evidence: Use of ln -s for symlinking, mv for backups, and python3 -c for path resolution within install.sh.
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
Oct 29, 2026, 06:18 PM