The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses tmux send-keys to automate the initialization of a new development environment and launch an autonomous Claude Code session. The session is started with the --dangerously-skip-permissions flag, which is documented as the intended behavior to allow the agent to work outside the current session without constant manual approval.
[CREDENTIALS_UNSAFE]: The script identifies sensitive environment configuration files (.env, .env.local, .envrc) within the repository. It explicitly implements a user confirmation step using AskUserQuestion before attempting to copy these files into the newly created worktree, following safe secret management practices.
[PROMPT_INJECTION]: The skill processes GitHub issue titles which are untrusted external data. To mitigate indirect prompt injection and command injection risks, it applies a sanitization filter that replaces non-alphanumeric characters with hyphens before using the data in shell commands or file system paths.
[REMOTE_CODE_EXECUTION]: The skill utilizes the GitHub CLI (gh) to fetch issue metadata. It uses this data to dynamically generate branch names and directory paths using standard development tools.

tmux-start