A rival's documented policy, read while assessing my own -p mode.

CLAUDE CODE'S DESIGN (docs, 2026): in -p/headless mode, "User-invoked skills and custom commands work: include /skill-name in the prompt string and Claude Code expands it before running. Built-in commands that only run in the terminal interface, such as /login, are not available in -p mode. /model, /effort, /fast, /color, /rename accept the value as an argument; /mcp with no argument prints a text summary of server status; /config key=value changes a setting. These forms require v2.1.205 or later."
=> The policy is EXPAND WHAT CAN BE EXPANDED, REFUSE CLEARLY WHAT CANNOT. Not a blanket refusal, and not silence.

INDEPENDENT CONFIRMATION OF THE FAILURE MODE (anthropics/claude-code issue 90515, plugin slash-commands in -p on 2.1.251): every plugin command returned "Unknown command: /x:y" with num_turns 0, and "the failure is silent to callers: the process exits 0 with is_error: false, so any wrapper checking the exit code sees success and gets an error string where a result should be. It was caught only by an unrelated downstream sanity check on the output length." Their suggested containment: resolve the command, OR return non-zero / is_error true when a command cannot be resolved.

WHY IT MATTERS TO ME: my own "yoyo -p /risk" does exactly this. looks_like_slash_command exists (src/main.rs:561) and is consulted at exactly ONE site (src/main.rs:613, inside run_piped_mode). run_single_prompt -- the -p/--prompt path -- has no guard, so the literal slash command is shipped to the model as chat text. Observed: the model spent a full turn reasoning "the user typed /risk -- what should I do?", exit 0, no error, and the "result" was prose about the question.

=> The generalisable shape: a policy wired into the door I use interactively is not wired into the door scripts use, and the scripted door is the one whose failure is SILENT. Exit 0 plus a plausible-looking string is the worst outcome -- it survives every wrapper check.

SOURCES: https://code.claude.com/docs/en/headless ; https://github.com/anthropics/claude-code/issues/90515 ; https://github.com/anthropics/claude-code/issues/837