GEO-first · SEO-active · Protocol-first
GEOHub organizes GEO research, diagnosis, content, measurement, and one-line SEO into a routable, reproducible, auditable super skill, with one entry point across six active capabilities
Six active skills share one Registry, Evidence Ledger, and artifact protocol
It turns natural-language requests into the smallest executable skill or a stable workflow, then records every run with evidence status, a quality report, and a manifest
GEOHub turns natural-language requests into five capability entry points:discover, diagnose, content, measure, and one-line SEO. The unified entry resolves intent, the Research Kernel governs scientific boundaries, and the Artifact Bus supports delivery and replay
A single Registry declares skill status, intents, entries, and outputs. The Router compiles its Action and Intent indexes from that source
Inputs pass strict schema, length, and file-boundary validation. Equivalent valid inputs produce stable structures without implicit web research
Factual claims link to evidence IDs. Missing sources remain unverified, source_gap, or blocked-by-evidence
Fifty-four papers and two datasets inform 11 runtime principles with explicit correlation, benchmark, causal, and governance status
The complete file set is generated and validated in a hidden staging directory, then published in one operation so failed runs remain invisible
If optional DOCX or PDF rendering fails, Markdown, JSON, and HTML remain available while the manifest records missing dependencies and renderer errors
The architecture separates natural-language selection, deterministic execution, research constraints, and delivery protocols while keeping shared data structures compatible
Inputs first pass character and byte budgets, Unicode normalization, quote-span detection, and negation-scope parsing. Planned capabilities always remain non-runnable
Each skill can be installed and executed independently or selected through the geo Router. The table summarizes inputs, core processing, and primary outputs
Scenarios are derived from the Registry, two stable DAGs, seven Content modes, and one-line SEO work modes. Every scenario traces to an active entry or an explicit planned state
Generate learn, compare, evaluate, and act questions from a product, service, or category seed
Organize FAQ, explainer, comparison, and landing-page opportunities by audience, scenario, and evidence gap
Check fact coverage for identity, offering, audience, differentiation, proof, and contact
Check discoverability, structure, extractability, evidence, authority, and freshness proxy signals
Generate compliant titles, definitions, steps, misconceptions, FAQ, and source sections
Generate neutral comparisons from aligned evidence or transparent rankings from complete score matrices
Design summaries, FAQ, tables, semantic HTML, schema candidates, CMS fields, and acceptance checklists
Preserve source fact boundaries while improving structure and expression. Unlinked facts remain unverified
Summarize answer rate, citation rate, conditional citation rate, missing answers, exclusion reasons, and uncertainty intervals
Recognize technical audits, keyword-to-page mapping, international SEO, ecommerce, and experiment modes from natural language
Stratify by page, query, market, device, and time while recording migration maps, validation, and rollback boundaries
Discover audience questions, then audit how well the brand, site, or page addresses them
Generate a query and opportunity map, then turn selected opportunities into titles, explainers, blueprints, or articles
Return required inputs, the nearest active capability, and the closest deliverable artifact
These examples come from Registry intents and the fixed routing contract. Add your topic, audience, URL, or evidence file after copying
The Artifact Bus places input snapshots, structured results, human-readable reports, the Evidence Ledger, research boundaries, and quality status in one delivery unit
runs/run-<id>/ ├── input/ │ ├── geo-brief.json │ └── sources/... ├── primary-result.json ├── report.md ├── content.html ├── evidence-ledger.json ├── research-context.json ├── quality-report.json └── run-manifest.json # Added by each skill query-map.json opportunity-map.json diagnosis-funnel.json content-evidence-units.json measurement-report.json seo-plan.json
| Skill | Core JSON | Human-readable result | Evidence and quality | Special artifacts |
|---|---|---|---|---|
| geo-discover | query-map、opportunity-map | Structured opportunity summary | ledger、research-context、quality、manifest | Four task forms with parent-query lineage |
| geo-diagnose | diagnosis、diagnosis-funnel | report.md | ledger、research-context、quality、manifest | HTML snapshots、remediation maps |
| geo-content | content-spec、content、evidence-units | content.md、content.html | ledger、research-context、quality、manifest | Optional DOCX / PDF |
| geo-measure | measurement-report | report.md | ledger、research-context、quality、manifest | Denominators, intervals, and platform strata |
| seo | seo-plan | report.md | ledger、quality、manifest | Work mode, evidence gaps, implementation authority, and rollback boundaries |
Fifty-four papers and two datasets from a pinned research commit were audited individually. Runtime uses a compact Registry and emits research-context.json for every execution
The local 0.3.1 snapshot passes the full test suite, fixed evaluations, cross-Python checks, deterministic packaging, isolated installation, and yao-meta Library gates