grok-geo

# grok-geo Skill > **Pattern: Pipeline + Inversion + Reviewer** > This skill enforces a strict multi-step pipeline with gate conditions. > It interviews the user for missing inputs before acting (Inversion). > It runs a quality review checklist before finalizing the report (Reviewer). ## Objective Produce a traceable AI-search/GEO audit for one brand using current web search, deterministic metric calculation, and evidence-backed recommendations. ## Required tools - web_search - shell If web_search is unavailable, switch to OFFLINE_IMPORT mode. Never fabricate search results or citations. ## Required input Minimum: - brand_name - website - industry - target_customer Recommended: - target_region - competitors - brand_aliases - products - known_facts - forbidden_claims ## Operating modes - **quick**: 10 questions, 1 query per question, 60-second snapshot - **standard**: 30 questions, up to 2 variants, full diagnostic - **offline_import**: analyze provided search results without new web searches ## Paths - Skill root: directory containing this `SKILL.md` - Scripts: `scripts/` - Default run base (hosted): `/mnt/data/geo-audit-runs` - Local override: environment variable `GEO_AUDIT_RUNS_DIR` or `./geo-audit-runs` - Python: use the runtime interpreter (`python3` / `python`) --- ## Phase 0 — Input Collection (Inversion Pattern) **DO NOT start the audit until all required inputs are confirmed.** If the user provides a partial input, ask for missing fields in this or...