-p / --print Flag

The -p / --print flag switches Claude Code from its default interactive REPL into a single-shot, headless mode. Pass it a prompt string, and Claude Code processes that prompt, writes the response to stdout, then exits with a zero status code on success (or non-zero on error). No TTY is required; no interactive session is opened; the process lifecycle is entirely predictable.

This behavioral change is what makes Claude Code composable with the rest of the shell ecosystem. A CI step, a cron job, a Git hook, or a Make target can invoke claude -p '...' and treat it like any other subprocess: capture stdout, inspect the exit code, pipe the output to the next stage. Without --print, Claude Code would hang waiting for interactive input in environments that have no terminal — a failure mode that can be surprisingly hard to diagnose when the process is wrapped inside a job runner.

The flag pairs naturally with --output-format to control how the response is structured. --output-format text (the default) is appropriate for human-readable summaries in GitHub Actions logs. --output-format json emits a structured envelope that downstream tooling can parse without fragile regex matching. --output-format stream-json emits newline-delimited JSON events as the model generates them, which is useful for long-running tasks where you want to update a progress display in real time rather than wait for the full response.

From a security standpoint, --print combined with a scoped allowed-tools configuration gives you an auditable, reproducible invocation. Every CI run with the same flags and the same prompt produces a traceable artifact: what Claude was asked, what tools it was permitted, and exactly what it returned. This is a meaningful improvement over ad-hoc scripting because the model's behavior is scoped and logged.

A common mistake is omitting --print in a shell script and then wondering why the script hangs. Another is forgetting that --print still runs the full model inference — it is not a "fast path" that skips reasoning. It changes the interaction model (non-interactive) not the compute model. Latency in --print mode is the same as latency in interactive mode for an equivalent prompt.

For exam scenarios, remember that --print is also required when you want to chain Claude Code inside a shell pipeline: claude -p 'Summarize this PR' | jq '.summary' only works because --print ensures Claude Code terminates after producing its output rather than entering an interactive loop. The CI/CD Integration with Claude Code concept page covers the full pattern — --print combined with --output-format and --json-schema — that the exam uses to test practical automation knowledge.

Exam questions typically present a CI/CD scenario — a GitHub Actions YAML, a shell script, or a Makefile — and ask which flag or combination of flags makes Claude Code usable in that context. Expect to distinguish --print (non-interactive execution) from --output-format (response structure) and --json-schema (output validation). You may also be asked to identify why a script that omits --print hangs in a headless environment.