The Agent Skills Directory

[COMMAND_EXECUTION]: The script jules-init.py generates and executes shell commands to detect project structure, read package.json, mise.toml, and docker-compose files. It also programmatically writes and sets execution permissions (chmod 755) for a new shell script scripts/jules-setup.sh.
[PRIVILEGE_ESCALATION]: The generated scripts/jules-setup.sh contains multiple sudo commands for installing system packages (apt-get install), managing services (systemctl start), and performing administrative database operations (sudo -u postgres psql). This assumes the environment allows passwordless sudo or requires the user to grant elevated privileges to an automated script.
[CREDENTIALS_UNSAFE]: The jules-init.py utility generates hardcoded database credentials based on the repository name and writes them directly into the generated jules-setup.sh file. Specifically, it creates roles and users with passwords identical to the username (e.g., CREATE ROLE tiller WITH LOGIN PASSWORD 'tiller').
[INDIRECT_PROMPT_INJECTION]: The skill automatically appends setup instructions to AGENTS.md or README.md. These instructions direct future AI agents to execute the generated scripts/jules-setup.sh script. This creates a surface where a malicious repository could include a pre-existing jules-setup.sh or trigger the generation of one that performs unauthorized actions when the next agent follows these instructions.
[COMMAND_EXECUTION]: The script modifies mise.toml by searching for and replacing shell command strings with more complex logic. This involves regex-based patching of configuration files that define execution hooks for the mise tool.

jules-config