Skip to content

troupe bench --live asks before it spends, at a terminal only, and --json and --compare take a value or none

TUI decision 141 · accepted · 2026-10-04

Root Decision 773. The plan and its cap go to standard error, so --json on standard output stays JSON. The question is asked only where troupe config asks one, with standard input and output a terminal, and the same way: on Windows key by key (Troupe.CLI.Prompt), since the binary's VM there has no reader on standard input (Decision 128), so a cooked read never returned, and the first installed run hung on it. y or yes runs it; anything else, or no terminal to ask in (a script, a pipe), runs nothing and exits 2; --yes answers beforehand. --repeat N, --model M and --yes belong to --live and are refused without it. --json stays the switch it was (troupe bench --json prints the report) and takes a file as well, --json FILE writing the report there and printing the table; --compare alone compares with the last bench and --compare REF with a version or a model. OptionParser has no switch with an optional value, so for troupe bench only, the word after either, unless it is a flag, is taken as its value before the parser runs; anywhere else --json is the plain switch. --md FILE writes the table to a file, live or offline. The TUI calls only Troupe.Bench (plan/1, describe_plan/1, question/1, live/2, compare/1), the door Decision 140 opened. Proof: test/troupe/bench_cli_test.exs, the question played through Bench.run/2's :ask, and with none, under capture_io, no terminal.

Governs: clients/tui/lib/troupe/cli.ex, clients/tui/lib/troupe/cli/bench.ex