| name | grok-fan-out |
| description | Run a multi-angle Grok analysis in a single call — Grok analyzes the task from several expert angles (security, review, research, tests) and synthesizes one consolidated answer with a section per angle plus a verdict. Use when the user wants a thorough, multi-perspective audit/review/analysis of something ("find everything wrong with…", "review this from all angles", "deep multi-angle look at…") that benefits from more coverage than a single-shot question. Read-only by default. |
Grok — Fan-Out (Multi-Angle Analysis)
Use this skill when one angle isn't enough — when the user wants something looked at from multiple specialties at once (correctness, security, tests, architecture, research) and a single consolidated verdict. The plugin runs one Grok analysis per angle in parallel, then a synthesis call that reconciles them — so you get broad coverage and real depth from a single command (deeper than a plain grok-ask). It makes several real Grok calls, so it's heavier — prefer it for substantial, multi-faceted tasks.
When to use
- "Audit / review this from every angle", "find everything wrong with X", "do a thorough multi-perspective analysis".
- A question that genuinely spans specialties: e.g. "is this auth module correct, secure, and well-tested?".
- You want a deeper second opinion than
grok-ask gives, without orchestrating multiple calls yourself.
Do not use this skill for:
- A simple one-off question →
grok-ask.
- Pure source-grounded research on a topic →
grok-research.
- Long-form coding / fixing / refactoring →
grok-rescue.
- Image generation →
grok-imagine.
How to invoke
Run exactly one bash command. Default to read-only (the subagents analyze; they don't edit):
node "${CLAUDE_PLUGIN_ROOT}/scripts/companion.mjs" fan-out "<the user's task, properly quoted>"
CLAUDE_PLUGIN_ROOT is set by both Claude Code and Codex CLI.
Choosing the angles
- Default angles (no flag):
researcher, reviewer, security-auditor, test-writer.
--personas a,b,c — pick a subset of the built-in angle names (researcher, reviewer, security-auditor, test-writer, implementer, design-doc-writer, design-doc-reviewer). Keep it to 3–4 angles unless the user explicitly asks for more — each angle adds analysis depth (and tokens). The plugin caps the total.--agents-json '<json array>' — advanced: name custom angles via a JSON array (each object needs a name). Only use when the built-in angle names don't fit the task.
Other flags
--write — only if the user explicitly asked the fan-out to make changes. Requires GROK_PLUGIN_ALLOW_WRITE=1 (the helper refuses otherwise — surface the refusal verbatim).--no-web-fetch — opt out of the default-on web_fetch grounding.--model <name> / --timeout <duration> — as for the other Grok commands.
Output rules
- Return the companion's stdout verbatim — it is Grok's consolidated multi-angle report (per-angle sections + a consolidated verdict). Do not paraphrase or re-summarize.
- If stderr contains a
[hint: ...] line about auth, surface it and direct the user to grok login / XAI_API_KEY.
- Fan-out is heavier than a single ask: prefer it for substantial, multi-faceted tasks, and don't fan out trivial questions.
