Claude Code One-Shot CLI

Problem

Automation workflows often stall on interactive confirmation, PTY dependency, or background tasks that are hard to observe. Claude Code CLI is capable, but when embedded directly in scripts, permission prompts, foreground blocking, and directory drift can prevent unattended execution. Script integrations especially need predictable output, a controlled workdir, and an execution path that does not depend on manual confirmation.

How it works

The skill wraps Claude Code as a one-shot coding agent. Its core requirement is to always use --print and --permission-mode bypassPermissions, which avoids confirmation prompts and keeps the output non-interactive. Tasks can run in the current directory or be pinned with workdir. Short tasks are suited to foreground execution; longer tasks can be run in the background and checked with process action:log. The documentation stresses not using PTY for Claude Code, because Claude Code does not require PTY and PTY can break the flow.

Boundaries

It is best for well-scoped engineering tasks such as feature building, refactoring, debugging, testing, and code review, but prompts must be specific. bypassPermissions reduces interaction friction while expanding automatic execution scope, so it should not be run casually in untrusted directories. Background tasks without monitoring can appear complete while still lacking verification.