feat(seo-skills): multi-backend Data Source Selection (#7)

Replaces single-vendor (Ahrefs-only) tool defaults with a per-task
backend menu across all 14 SEO skills. Each skill now lists every
capable MCP in allowed-tools and documents how to pick between
Semrush, Ahrefs, OurSEO Agent (CLI + MCP), DataForSEO, and GSC
in its SKILL.md Data Source Selection section.

Tool stubs (~40 new files) populated per skill with capability
deltas, call patterns, and explicit "not for this skill when"
callouts so the menu is self-correcting.

Skills affected: 19-keyword-strategy, 20-serp-analysis,
21-position-tracking, 22-link-building, 23-content-strategy,
24-ecommerce, 25-kpi-framework, 26-international, 27-ai-visibility,
28-knowledge-graph, 31-competitor-intel, 32-crawl-budget,
33-migration-planner, 34-reporting-dashboard.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

custom-skills/25-seo-kpi-framework/desktop/SKILL.md

@@ -21,19 +21,80 @@ Aggregate SEO KPIs across all dimensions into a unified dashboard. Establish bas
 5. **Performance Reporting** - Period-over-period comparison with executive summary
 6. **Tactical Breakdown** - Actionable next steps per dimension
 
-## MCP Tool Usage
+## Data Source Selection
 
-### SEO Data
+KPI framework is an **aggregator** — it pulls one metric per dimension from the best backend for that metric, then composites them. **Pick per metric, not per skill invocation.** Don't pull every metric from every backend.
+
+### Per-metric backend defaults
+
+| KPI dimension | Default backend | Alternates | Notes |
+|---|---|---|---|
+| **Organic traffic + traffic value** | Semrush `overview_research` | Ahrefs `site-explorer-metrics` | Modelled estimates — pick one per audit and document it. |
+| **Visibility / ranking distribution** | Semrush `tracking_research` (when project exists), else Ahrefs `rank-tracker-*` | OurSEO `check_serp` for spot positions | First-party alternative: GSC position data. |
+| **Backlinks / DR** | Ahrefs `site-explorer-domain-rating` + `-backlinks-stats` | Semrush `backlink_research` | Ahrefs has the strongest backlink graph. |
+| **Technical health** | OurSEO `our audit tech` + `our analyze mysql-batch` | Ahrefs `site-audit-issues`, Semrush `siteaudit_research` | OurSEO is the deepest because it owns the crawl + fix engine. |
+| **Indexed pages** | OurSEO `our research google index` / `mcp__ourseo__check_index` | GSC index coverage report | First-party best. |
+| **Content freshness** | OurSEO `crawl_website` (extracts last-modified) | Ahrefs `site-explorer-pages-history` | |
+| **First-party clicks / impressions / CTR** | **GSC** via `our research search-console` | Ahrefs `gsc-*` if Ahrefs project connected | **Required for accurate KPI** — Google's own data, not modelled. |
+| **GA4 / on-site engagement** | OurSEO CLI: `our research ga4 *` | (none equivalent in Semrush/Ahrefs at the property level) | Sessions, bounce, conversion. |
+| **Local visibility (GBP)** | OurSEO `our collect gbp *` + `our audit local` | (Semrush local listings only in US/EU) | First-party Google Business Profile data. |
+
+### How to pick
+
+1. **User named a backend for a specific metric** → use it for that metric.
+2. **User preference memory** — read `feedback_seo_tool_preferences.md`; honor per-task defaults.
+3. **Apply per-metric defaults from the table above.** Default to first-party (GSC, GA4, GBP) whenever it's available — every modelled estimate weakens the composite.
+4. **Be consistent across reporting periods.** If the prior baseline used Semrush traffic value, use Semrush traffic value this run too — switching mid-stream breaks the trend.
+5. **Document every metric's source in the report Overview.**
+
+### Backend call patterns
+
+**Semrush MCP (default traffic + visibility):**
 ```
-our-seo-agent CLI: Primary metrics source (future); use --input for pre-fetched JSON
-WebSearch / WebFetch: Supplementary metrics data
+mcp__semrush__overview_research(query="<domain>", database="us")
+mcp__semrush__organic_research(query="<domain>", database="us")
+mcp__semrush__tracking_research(query="<keyword>", database="us")
 ```
 
-### Notion for Report Storage
+**Ahrefs MCP (backlinks + traffic-value alternate):**
 ```
-mcp__notion__*: Save reports to SEO Audit Log database
+mcp__ahrefs__site-explorer-metrics(target="<domain>")
+mcp__ahrefs__site-explorer-domain-rating(target="<domain>")
+mcp__ahrefs__site-explorer-domain-rating-history(target="<domain>", history="weekly")
+mcp__ahrefs__site-explorer-backlinks-stats(target="<domain>")
 ```
 
+**OurSEO (technical + indexation + first-party):**
+```bash
+our audit tech https://<domain>
+our analyze mysql-batch --session <id>
+our research google index --domain <domain>
+our research search-console queries --site sc-domain:<domain> --days 28
+our research ga4 traffic --property-id <id>
+our audit local https://<domain> --gbp-profile <client>
+```
+
+**OurSEO MCP (Claude Desktop alternate):**
+```
+mcp__ourseo__check_index(domain="<domain>")
+mcp__ourseo__audit_page(url="<url>", audit_type="tech")
+```
+
+### Reporting rule
+
+Every KPI report's **Overview** section MUST include a "Sources" subsection listing the data source per metric. Example:
+
+```markdown
+### Sources
+- Organic traffic: Semrush overview_research (database=us)
+- Backlinks / DR: Ahrefs site-explorer-domain-rating
+- Indexed pages: OurSEO check_index
+- Clicks / Impressions: GSC (28d)
+- GBP visibility: OurSEO collect gbp (profile=client)
+```
+
+This is non-negotiable — period-over-period KPI comparisons are meaningless without per-metric source attribution.
+
 ## Workflow
 
 ### 1. KPI Aggregation