geo-skills

A free, open-source GEO (Generative Engine Optimization) skill pack. Built as portable Agent Skills - runs in Claude Code, Codex, and other agents that read skill files.

If this saves you a paid dashboard or a week of GEO research, star the repo so more operators can find it.

Traditional SEO ranks you in Google's blue links. GEO gets you quoted inside the AI answer - by ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews. This pack scores how citable your site is to those engines and writes you a prioritized fix list. No dashboard subscription, no paid API required. It runs locally in Claude Code.

What it does

Point the orchestrator at a URL. It crawls your site, runs a panel of specialist subagents, and hands you a single GEO Score (0-100), weighted across 6 categories:

Then it ranks every issue by severity (critical first, quick wins flagged) and gives you a 30-day action plan.

Paste-ready search intelligence prompts

The optional prompt-packs/dataforseo-intelligence pack turns live search and AI-visibility data into decision-ready reports. It includes site explorer, keyword research, competitor gap, technical audit, AI visibility, community-demand, and owner-intent mapping prompts.

These are not generic "write me an SEO report" prompts. Each one has a spend cap, coverage receipt, evidence-state labels, fail-closed missing-data rules, and an exact handoff. The owner-intent prompt is the front door to TokenMax: it decides whether to upgrade, create, consolidate, or hold a page before generation starts.

What's in the pack

GEO core (the flagship):

Supporting SEO skills:

programmatic-seo · schema-markup · competitor-alternatives · seo-audit · content-strategy · site-architecture

Subagents (in agents/, used by the orchestrator): geo-ai-visibility, geo-content, geo-platform-analysis, geo-schema, geo-technical.

Compatibility

These are standard Agent Skills - each one is a SKILL.md instruction document plus a rubric and output format. That makes them portable:

The individual skills (citability, crawlers, schema, content, and the rest) are plain instruction docs - any capable agent that can fetch a URL and write a file can run them. The one Claude Code-native piece is the geo orchestrator's parallel subagent fan-out (it launches 5 specialists at once). On other runtimes you invoke the skills directly, or the agent runs them in sequence - same checks, same output. The allowed-tools frontmatter is advisory and is safely ignored by agents that don't use it.

Install

See INSTALL.md. Short version: copy skills/ into your agent's skills directory (~/.claude/skills/ for Claude Code, ~/.codex/skills/ for Codex) and, on Claude Code, agents/ into ~/.claude/agents/. Then:

/geo audit https://yoursite.com

Agent-native CLI

Install the command surface from a clone:

python -m pip install -e ".[web]"
geoskills --json doctor
geoskills skills list --json
geoskills prompts list --json

The CLI exposes deterministic discovery and validation without pretending every GEO workflow is a local script. It can validate all bundled Agent Skills, inspect a page, score citability, and validate the versioned owner-intent JSON handoff consumed by TokenMax:

geoskills --json validate
geoskills inspect https://example.com --mode full --json
geoskills citability https://example.com --json
geoskills owner-map validate owner-intent-map.json --json

Every CLI command is read-only. Full LLM audits remain Agent Skills so the runtime, model, and any paid connector stay explicit.

The command contract follows the agent-native principles demonstrated by CLI-Anything: self-describing help, machine-readable output, installed-command tests, and real backend behavior. This implementation is purpose-built for GEO rather than generated from their harness.

Free-first

Every skill in this pack works with no paid API. Where a third-party data tool could help (rank trackers, backlink suites), it is always marked optional and the skill degrades gracefully without it. Research figures cited in the skills come from published studies and are attributed inline.

The optional DataForSEO prompt pack is intentionally separate from the free core. It uses DataForSEO's paid-per-use API through the official MCP server, requires an active account, and tells the agent to stop at the operator's task/row cap. No DataForSEO credentials belong in a prompt or repository file.

The heavy-artillery skills are token-max-factory, which drives the free, MIT Archon CLI and engine repo, and tokenmax-fleet-orchestrator, which sequences consumer-owned site/deploy adapters. They use the coding-agent subscription already running the pack (codex by default, claude opt-in). Still no SEO APIs, no data vendors, no dashboards.

License

MIT. See LICENSE. Built by SmokeDev. PRs welcome - see CONTRIBUTING.md. If the toolkit supports published work, use GitHub's Cite this repository control backed by CITATION.cff.