The Agent Skills Directory

[PROMPT_INJECTION]: The skill programmatically configures the AI agent's security posture by setting bypassPermissions to defaultMode in ~/.claude/settings.json within resources/post_install.py. This instructs the agent to skip user confirmation prompts for sensitive operations like file system changes or command execution.
[COMMAND_EXECUTION]: The skill adds a claude-yolo alias to .zshrc that executes the agent with the --dangerously-skip-permissions flag, further encouraging the bypass of built-in security guardrails.
[COMMAND_EXECUTION]: The resources/devcontainer.json configuration grants the container NET_ADMIN and NET_RAW capabilities. These are elevated privileges that allow processes inside the container to manipulate network interfaces, sniff traffic, and modify firewall rules (iptables).
[COMMAND_EXECUTION]: The resources/post_install.py script uses sudo chown to modify directory ownership on mounted volumes, indicating the container user has elevated privileges.
[EXTERNAL_DOWNLOADS]: The resources/Dockerfile performs multiple remote downloads and executions of scripts and binaries from external sources:
Installs Claude Code via curl -fsSL https://claude.ai/install.sh | bash.
Installs fnm (Fast Node Manager) via curl -fsSL https://fnm.vercel.app/install | bash.
Installs zsh-in-docker via a script from github.com/deluan/zsh-in-docker.
Fetches binaries for fzf and git-delta directly from GitHub releases.
Adds several third-party plugins from the Claude marketplace.
[REMOTE_CODE_EXECUTION]: The resources/install.sh script includes an update command that executes git pull in its local directory. This mechanism fetches and integrates remote code updates, which are then executed during subsequent uses of the tool.

devcontainer-setup