gate

One call runs every gate command and reads each real exit code. A command that exits 0 but prints failures (Tests 1 failed, # fail 2) is marked failed. Trust the runner, not a summary line.

"Stop" means: end your turn with the report and wait.

1. Run node "${CLAUDE_PLUGIN_ROOT}/scripts/gate/run.mjs" exactly like that, forward slashes included, once, from any folder in the project; it prints every command's exit code itself, so never append echo $?, pipes or cd. Re-run only after you changed code, or once with --plan <path> when a row says no plan for this branch. Add --plan <path> when you know the task's plan (relative paths are relative to the project root; without it the runner looks for the plan of the current branch). Add --only <name,name> only when asked for some of the commands. The first stderr line, nauraiz-dev gate: project root <dir>, says where it ran.
2. A nauraiz-dev: line printed before the table names a config problem: quote it in the report whatever the exit code. Read the exit code:
   • 0: every command passed.
   • 1: at least one failed: a non-zero exit, or exit 0 but output reports failures. A row failed with no plan for this branch (pass --plan) was not run: if a plan for this task exists, run again with --plan <path>; otherwise report it failed.
   • 2: quote every nauraiz-dev: line it printed first. If any line other than no gate configured … names a config file, the config failed to load: say the file needs fixing, not that there is no gate, and stop. Otherwise no gate is configured: say so, point to /nauraiz-dev:init (it detects the gate commands), and stop.
   • 3: the runner itself failed (nauraiz-dev: gate runner failed: …). Quote its output and stop. Any other non-zero exit: the same.
3. Paste the runner's output verbatim (the table and every block), then one line: Gate: pass or Gate: FAIL (<names>).
4. Never re-run a gate command by hand, pipe it or grep it, unless the runner reported that command failed; then run only that one, only to find the cause. A by-hand run is never the gate result: after a fix, run the runner again.
5. Never call anything green that failed or did not run. Called on its own: stop after the report. Called from another skill: hand the result back to it.