GEO SEO Hub

Version 0.5.0 · Experimental · GEO-first · SEO-ready

GEO SEO Hub is an open, protocol-first agent skill hub with active GEO workflows and an SEO-ready shared foundation. Version 0.5.0 turns generative engine optimization work into auditable, reusable runs through a registry-driven control plane, Artifact Bus, evaluation lab, discovery, diagnosis, content, measurement, strategy optimization, knowledge governance, JSON Schema contracts, and seven active Skills.

The skills are Library-engineered packages while product behavior remains Experimental. maturity_tier=library describes packaging rigor and does not claim production outcome quality.

Current scope

• geo: routes Chinese and English requests through the registry and reports unavailable routes honestly.
• geo-discover: converts a validated GEO brief into a deterministic query map, opportunity map, evidence ledger, run manifest, and quality report.
• geo-diagnose: evaluates explicit brand, site, or page sources and emits a structured diagnosis, deterministic report, evidence ledger, remediation query map, opportunity map, quality report, and run manifest.
• geo-content: creates evidence-lined titles, explainers, comparisons, rankings, page blueprints, refinements, and article-friendly artifacts as JSON, Markdown, and standalone HTML; DOCX/PDF are optional render layers.
• geo-measure: computes reproducible visibility, mention, source-inclusion, citation-share, and panel-coverage metrics from approved offline engine-observation bundles.
• geo-strategy: produces bounded intervention candidates, fidelity checks, experiment plans, external publication handoffs, and positive-only strategy memory.
• geo-knowledge: builds source-lined entity graphs, preserves conflicting facts, supports local/global queries, and replaces changed source-hash contributions.
• geo-publish: remains a visible roadmap route with planned status and no execution authority.

The resolver keeps single-intent routing minimal and exposes two exact multi-stage DAGs: brand-baseline-lite (discover → diagnose) and content-campaign (discover → content). Lexical routing keeps production authority; semantic scoring runs in shadow mode and records disagreements. Durable workflow state supports checkpoints, retry, abort, approval, and explicit external publication/observation waits. The planned publish route returns its closest active suggestion and required inputs. Route requests are bounded to 8,000 characters and 16,384 UTF-8 bytes. See skills/RESOLVER.md and docs/architecture.md.

No connector, live platform sampling, search volume, ranking, or conversion data is inferred. Missing evidence remains explicit in generated artifacts. Every executable run includes run-lineage.json with artifact hashes and allowlisted operational metadata. Run payloads stay out of adoption and drift aggregation.

GEO and SEO boundary

GEO SEO Hub treats GEO and SEO as related capability domains with a shared search foundation. Query intent, source evidence, brand facts, entity structure, content specifications, site parsing, quality reports, and Artifact Bus contracts can serve both domains.

Version 0.5.0 ships executable GEO discovery, diagnosis, content, observation-bundle measurement, strategy planning, knowledge governance, and routing. Strategy execution ends at an external publication handoff and cannot claim effect without verified post-window measurement. Dedicated SEO workflows and outcome claims for technical SEO, SERP and keyword data, indexation, Core Web Vitals, internal linking, Search Console, and traffic measurement remain future Registry additions.

Version 0.5 retains geo-seo-hub as the distribution and CLI namespace, geo_seo_hub as the Python module, and share/geo-seo-hub as the installed data root. Existing geo-* Skill IDs, public imports, CLI contracts, legacy execution defaults, and Artifact Bus protocol 1.0.0 remain stable. See docs/migration-0.5.md and CHANGELOG.md.

Quick start

Supported Python range: 3.11-3.14.

git clone https://github.com/yaojingang/geo-seo-hub.git
cd geo-seo-hub
python3 -m venv .venv
.venv/bin/python -m pip install .
.venv/bin/geo-seo-hub --version
.venv/bin/geo-seo-hub route --text "帮我挖掘 AI 搜索问题"

For development and the full release gate:

.venv/bin/python -m pip install -e '.[dev]'
.venv/bin/python scripts/verify_all.py

Community packages:

python3 scripts/package.py --target all --channel community
python3 scripts/verify_packages.py
python3 scripts/install_simulation.py --target all

The eleven artifacts are a source ZIP, one unified single-Skill ZIP, seven provider Skill ZIPs, and Codex/Claude adapter ZIPs. Each ZIP supports pip install . from its extraction root. Unified and target adapters contain seven parseable provider entries and wrappers. Every community artifact is AGPL-3.0-only; commercial metadata remains inquiry_only. Version 0.5.0 has no verified GitHub Release assets in this source snapshot; build packages locally or run the attested release workflow. See docs/installation.md.

Version 0.2 removes the pre-release 0.1 runtime aliases. Recreate the environment before installing 0.2; the exact historical mapping remains in the migration ledger and third-party notice.

The CLI prints JSON. The --output value is a runs root; all seven provider executors write protocol 1.0.0 runs to <output>/<run-id>/ and return that actual run directory. Content runs never access the network, snapshot relative source files for offline replay, escape user text in standalone HTML, and keep optional renderer failures explicit. If DOCX/PDF dependencies are missing, core output succeeds and the run manifest records degraded plus missing_dependencies. Install .[render] to request DOCX/PDF support. Diagnose fetches only explicit public HTTP(S) canonical URLs without query strings, accepts HTML/XHTML, performs no crawl expansion, and snapshots successful pages for offline replay. Unavailable or unsupported sources remain gaps. Its scores do not represent live AI-platform recall, ranking, or citation share.

Release evidence is generated with scripts/generate_sbom.py, scripts/generate_provenance.py, scripts/verify_provenance.py, and scripts/render_production_readiness.py. Local provenance is explicitly unsigned. The current Production Readiness decision stays blocked while human review, live platform benchmarks, governed adoption, external strategy effect, production knowledge evaluation, legal review, and executed CI attestation remain missing evidence.

Data retention is preview-only by default. geo-seo-hub data-retention --runs-root <root> --apply-policy lists expired runs. Add --confirm to move them into a recoverable .geohub-trash batch. Recovery is explicit, and permanent purge requires a second confirmation after a seven-day grace period. See docs/data-governance.md.

Visual reports

• GEOHub 0.5.0 upgrade review provides the standalone Chinese HTML report with 17-module before-and-after scores, repair closure, deterministic gates, and remaining Production evidence.

License and governance

The open-source repository is licensed under AGPL-3.0-only. Commercial licensing is currently inquiry_only; see COMMERCIAL-LICENSING.md. The contributor agreement remains under legal review, so external code merges are paused.

make verify and python3 scripts/verify_all.py run the complete repository, evidence, test, evaluation, package, and isolated-install gates. make repo-verify runs the fast structural check. The gates use synthetic fixtures and require no external service or secret.

Copyright © 2026 姚金刚 / Yao.