colab-ssh-setup

Warn

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill installs the mighty-colab package from PyPI and the google-colab-cli tool from a GitHub repository (github.com/googlecolab/google-colab-cli.git). While the GitHub repository belongs to a well-known service (Google Colab), mighty-colab is an external third-party package provided by the author as a workaround for official tool bugs.
  • [PERSISTENCE]: The skill modifies the user's local ~/.ssh/config file to add permanent host aliases (colab-agent, colab-shell) and ProxyCommand directives. This modification persists across sessions and alters the local system's SSH behavior.
  • [COMMAND_EXECUTION]: The skill executes multiple shell commands to install tools (uv tool install), clone repositories (git clone), and manage remote Colab sessions (colab new, colab ssh, colab stop).
  • [INDIRECT_PROMPT_INJECTION]: The skill accepts a user-defined session name and interpolates it directly into shell commands and the SSH configuration file.
  • Ingestion points: The user-provided <session_name> (SKILL.md).
  • Boundary markers: No explicit delimiters or boundary markers are used in the command interpolation logic.
  • Capability inventory: Includes shell command execution and file-writing to the sensitive ~/.ssh/config file.
  • Sanitization: The skill relies on a natural language instruction to "Avoid spaces and shell-special characters" but lacks programmatic sanitization or validation, making it vulnerable if the agent fails to follow the constraint.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 28, 2026, 01:19 PM
Security Audit — agent-trust-hub — colab-ssh-setup