stop-gate

Consult — Stop Gate

Run a stop-gate-style review. The reviewer is asked: "is this turn done correctly, or are there issues remaining?" and produces VERDICT: PASS | NEEDS FIXES.

This command runs the same logic the optional Stop hook uses, so you can preview behavior before enabling the hook.

Behavior

node ${CLAUDE_PLUGIN_ROOT}/scripts/consult.mjs stop \
  --provider ${ARGUMENTS.provider:-zai}

Surface the verdict and any listed issues. If NEEDS FIXES, the underlying script also exits non-zero — but in manual invocation just show the result; don't treat it as a script failure.

Enabling the Stop hook

The Stop hook is opt-in. To enable it, add this to your project's .claude/settings.json (or use the global ~/.claude/settings.json):

{
  "hooks": {
    "Stop": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "node ${CLAUDE_PLUGIN_ROOT}/scripts/consult.mjs stop --provider zai"
      }]
    }]
  }
}

When VERDICT: NEEDS FIXES is returned, the script exits 2 and Claude Code surfaces the reviewer's notes back to the model, prompting another turn.

${CLAUDE_PLUGIN_ROOT} caveat: that variable is expanded for hooks defined in a plugin's own hooks/hooks.json, but it may not be substituted inside a user-authored .claude/settings.json. If the hook fails to find the script, replace ${CLAUDE_PLUGIN_ROOT} with the absolute path to the plugin's scripts/consult.mjs.

Privacy / consent: the stop gate sends the tail of your transcript and the current git diff to a third-party API (z.ai or Google). Only enable it on work you're comfortable sharing with that provider, and avoid it in repos containing secrets or regulated data. Redact sensitive content before enabling, or point --provider at a provider whose data policy you accept.