The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interpolates the parameter, derived from natural language $ARGUMENTS, directly into shell commands such as git worktree add, rm -rf, and cd. This pattern is vulnerable to command injection and path traversal if the user input contains shell metacharacters or relative path sequences.\n- [DATA_EXFILTRATION]: The skill accesses and copies sensitive files, including .env.local, service-account-key.json, and private keys (.certs/getatomi.local.key). While the files are copied locally, this behavior exposes sensitive credentials to the newly created environment.\n- [REMOTE_CODE_EXECUTION]: The skill executes direnv allow and mise trust, which explicitly bypass security prompts to allow the execution of project-local configuration scripts (.envrc, .mise.toml). This can lead to arbitrary code execution if the repository contains malicious configuration files. It also installs dependencies using pnpm i.

platform-worktree