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/23-seo-content-strategy/desktop/SKILL.md

@@ -18,25 +18,74 @@ Audit existing content performance, identify topic gaps vs competitors, map topi
 4. **Editorial Calendar** - Prioritized content creation schedule
 5. **Korean Content Patterns** - Naver Blog style, 후기, 추천 format analysis
 
-## MCP Tool Usage
+## Data Source Selection
 
-### SEO Data
+This skill can pull content + keyword + traffic data from multiple backends. **Pick one backend per data type per task** — content strategy spans multiple data classes, so you'll often combine 2 backends (e.g., one for keyword discovery + one for on-page inventory).
+
+| Backend | Best for | Notes |
+|---|---|---|
+| **Semrush MCP** (`mcp__semrush__*`) | **Default** for organic content discovery, keyword expansion, top pages by traffic | `organic_research`, `keyword_research`, `overview_research` → `get_report_schema` → `execute_report`. |
+| **Ahrefs MCP** (`mcp__ahrefs__*`) | Top-pages-by-traffic for competitor sites, content-decay candidates, Korean platform refdomains | `site-explorer-top-pages`, `site-explorer-pages-history`, `keywords-explorer-*`, `gsc-pages` (first-party). |
+| **OurSEO MCP** (`mcp__ourseo__*`) | **On-page content inventory** — what the target site itself publishes | `crawl_website` for content URLs + titles + H1s; `find_similar_pages` for topic clustering. Pair with a keyword backend for performance data. |
+| **OurSEO CLI** (`our keywords *`, `our serp *`) | DataForSEO under the hood for Korean batch keyword + competitor pulls | Claude Code only (Bash). Best for `--location 2410` Korean content gap work. |
+| **GSC** (via `our research search-console` or Ahrefs `gsc-*`) | **Cannibalization detection** — query × page where multiple URLs split impressions; first-party content performance | Only first-party source — required for accurate decay detection on the target site. |
+| **DataForSEO MCP** (`mcp__dfs-mcp__*`) | Fallback when `our` CLI isn't running | Same data as `our keywords *` / `our serp *`. |
+
+### How to pick
+
+1. **User named a backend explicitly** → use it.
+2. **User preference memory** — read `feedback_seo_tool_preferences.md`; honor the task-type default.
+3. **Task is content decay / cannibalization on the target site** → use **GSC** (first-party impression data is required).
+4. **Task is on-page content inventory** → use **OurSEO crawl_website**.
+5. **Default for keyword + competitor pulls**: Semrush MCP (English markets); OurSEO CLI (`--location 2410`) for Korean markets.
+6. **Still ambiguous + non-trivial** → ask once via `AskUserQuestion`.
+
+### Backend call patterns
+
+**Semrush MCP (default keyword/content discovery):**
 ```
-our-seo-agent CLI: Primary content/traffic data source (future); use --input for pre-fetched JSON
-WebSearch / WebFetch: Supplementary content data
+mcp__semrush__organic_research(query="<domain>", database="us")
+mcp__semrush__keyword_research(query="<seed>", database="us")
+mcp__semrush__get_report_schema(report_id="...")
+mcp__semrush__execute_report(report_id="...", params={...})
 ```
 
-### WebSearch for Content Research
+**Ahrefs MCP (top pages by traffic, decay candidates):**
 ```
-WebSearch: Research content topics and competitor strategies
-WebFetch: Analyze competitor page content and structure
+mcp__ahrefs__site-explorer-top-pages(target="<domain>", country="us", limit=100)
+mcp__ahrefs__site-explorer-pages-history(target="<domain>", history="monthly")
+mcp__ahrefs__keywords-explorer-overview(keyword="<seed>", country="us")
 ```
 
-### Notion for Report Storage
+**OurSEO MCP (on-page content inventory):**
 ```
-notion-create-pages: Save audit reports to SEO Audit Log
+mcp__ourseo__crawl_website(url="<target>", max_pages=200)
+mcp__ourseo__find_similar_pages(crawl_path="<path/to/crawl.json>", query="<topic>")
 ```
 
+**OurSEO CLI (Korean batch):**
+```bash
+our keywords ideas "<seed>" --location 2410 --limit 50
+our keywords for-site <competitor.com> --location 2410 --limit 100
+our serp ranked-keywords <domain> --location 2410 --limit 100
+```
+
+**GSC (cannibalization + decay):**
+```bash
+our research search-console combined --site sc-domain:<domain> --days 90
+# Group by query; flag queries where multiple pages share impressions.
+```
+
+### Common parameters
+
+| Concept | Semrush | Ahrefs | DataForSEO / `our` CLI |
+|---|---|---|---|
+| Korean market | `database="kr"` | `country="kr"` | `--location 2410` |
+| US market | `database="us"` | `country="us"` | `--location 2840` |
+| Japan | `database="jp"` | `country="jp"` | `--location 2392` |
+
+Always record the chosen data source(s) in the report **Overview** so future audits can compare like-for-like.
+
 ## Workflow
 
 ### 1. Content Audit

custom-skills/23-seo-content-strategy/desktop/skill.yaml

@@ -1,8 +1,16 @@
 name: seo-content-strategy
 description: |
   Content strategy and planning for SEO. Triggers: content audit, content strategy, content gap, topic clusters, content brief, editorial calendar, content decay.
+
+# Allowed tools list every backend the skill can pull content/keyword data from.
+# Per-task selection happens in SKILL.md > Data Source Selection — NOT here.
 allowed-tools:
-  - mcp__ahrefs__*
+  - mcp__semrush__*               # default for organic content + keyword discovery
+  - mcp__ahrefs__*                # site-explorer-top-pages, keywords-explorer
+  - mcp__claude_ai_Ahrefs__*      # Ahrefs (Claude.ai namespace)
+  - mcp__ourseo__*                # crawl_website for on-page content inventory
+  - mcp__dfs-mcp__*               # DataForSEO MCP fallback
+  - Bash                          # `our keywords *`, `our serp *` CLI (Claude Code only)
   - mcp__notion__*
   - WebSearch
   - WebFetch

custom-skills/23-seo-content-strategy/desktop/tools/ahrefs.md

@@ -1,15 +1,32 @@
-# Ahrefs
+# Ahrefs MCP
 
-> TODO: Document tool usage for this skill
+Use for **top-pages-by-traffic** views on competitor sites, **content decay** candidates on the target, and Korean platform refdomains. Namespace: `mcp__ahrefs__*` / `mcp__claude_ai_Ahrefs__*`.
 
-## Available Commands
+## Key endpoints
 
-- [ ] List commands
+- `site-explorer-top-pages` — pages ranked by traffic for any domain (target or competitor)
+- `site-explorer-pages-history` — trend in indexed pages over time (decay signal)
+- `site-explorer-pages-by-traffic` — sorted by current traffic
+- `site-explorer-pages-by-internal-links` — informs internal-linking opportunities
+- `keywords-explorer-overview` / `-related-terms` / `-matching-terms` — topic expansion
+- `gsc-pages` — first-party top pages when GSC is connected to the project
 
-## Configuration
+## Example
 
-- [ ] Add configuration details
+```
+mcp__ahrefs__site-explorer-top-pages(target="<competitor>", country="us", limit=100)
+mcp__ahrefs__site-explorer-pages-history(target="<target>", history="monthly")
+mcp__ahrefs__keywords-explorer-related-terms(keyword="<topic>", country="us", limit=50)
+```
 
-## Examples
+## Strengths for this skill
 
-- [ ] Add usage examples
+- Best ranked-by-traffic view of competitor content inventory.
+- Historical pages-history surfaces decaying content quickly.
+- Pairs well with OurSEO crawl (which gives the on-page side).
+
+## Not for this skill when
+
+- **On-page content inventory of the target** — OurSEO `crawl_website` is the ground-truth source.
+- **Cannibalization detection** on the target — GSC first-party data is required.
+- **Korean Naver content** — Ahrefs underrepresents Naver Blog/Cafe; use `our research naver *` and WebSearch.

custom-skills/23-seo-content-strategy/desktop/tools/gsc.md

@@ -0,0 +1,42 @@
+# Google Search Console
+
+**Required** for two content-strategy tasks that no other backend can do honestly:
+
+1. **Cannibalization detection** — query × page combos where multiple URLs share impressions.
+2. **Decay detection** — period-over-period impression decline on the target's existing pages.
+
+## CLI commands
+
+```bash
+# Top queries — what the site already ranks for
+our research search-console queries --site sc-domain:<domain> --days 28
+
+# Top pages by clicks/impressions
+our research search-console pages --site sc-domain:<domain> --days 28
+
+# Combined query × page — REQUIRED for cannibalization
+our research search-console combined --site sc-domain:<domain> --days 90
+```
+
+`sc-domain:` prefix for Domain-verified properties; URL-prefix properties use the plain URL.
+
+## Decay workflow
+
+1. Pull `pages` for the last 90 days.
+2. Pull `pages` for the same 90-day window from 90 days prior (or use cached prior runs).
+3. Diff impression / click counts per URL.
+4. Flag URLs with >30% impression decline as decay candidates.
+5. Cross-reference with `our collect crawl` output to check freshness signals (last-modified, content age).
+
+## Cannibalization workflow
+
+1. Pull `combined` over 90 days.
+2. Group by query; count distinct URLs per query.
+3. Flag queries where 2+ URLs each have >100 impressions — the site is competing with itself.
+
+## Not for this skill when
+
+- **Net-new content discovery** — GSC only knows what already ranks. Use Semrush/Ahrefs for gap discovery.
+- **Competitor performance** — single-site only.
+
+Gotcha: `our-claude-skills/custom-skills/15-seo-search-console/code/gotcha/gsc-cli-integration.md`.

custom-skills/23-seo-content-strategy/desktop/tools/ourseo.md

@@ -0,0 +1,39 @@
+# OurSEO Agent (CLI + MCP)
+
+The **on-page** half of content strategy work — what the target site itself publishes, in what shape, with what schema. Pair with a keyword backend (Semrush/Ahrefs) for performance data.
+
+## CLI commands (Claude Code)
+
+```bash
+# Crawl the target for content inventory (titles, H1s, meta, last-modified)
+our collect crawl https://<site> --max-pages 500
+
+# Korean batch keyword research for content gaps
+our keywords ideas "<seed>" --location 2410 --limit 50
+our keywords for-site <competitor.com> --location 2410 --limit 100
+our serp ranked-keywords <domain> --location 2410 --limit 100
+
+# Naver-specific content discovery
+our research naver keywords ideas "<seed>" --limit 30
+our research naver serp "<keyword>"
+```
+
+## MCP tools (Claude Desktop)
+
+```
+mcp__ourseo__crawl_website(url="<target>", max_pages=200)
+mcp__ourseo__find_similar_pages(crawl_path="<path/to/crawl.json>", query="<topic>")
+mcp__ourseo__search_knowledge_graph(query="<entity>")
+```
+
+## Strengths for this skill
+
+- **`crawl_website` is the ground-truth on-page content inventory.** No vendor estimates — just what the site publishes.
+- **`find_similar_pages`** does semantic clustering on a prior crawl (Vertex AI embeddings) — useful for topic-cluster building and cannibalization candidates.
+- Korean Naver content discovery via `our research naver *` covers a market Semrush/Ahrefs don't reach.
+
+## Not for this skill when
+
+- **Search volume / KD per keyword** — Semrush MCP is the cleanest source.
+- **Top pages by traffic on competitor sites** — Ahrefs `site-explorer-top-pages`.
+- **First-party content performance** — GSC `pages` / `combined` reports.

custom-skills/23-seo-content-strategy/desktop/tools/semrush.md

@@ -0,0 +1,34 @@
+# Semrush MCP
+
+**Default** for organic content discovery and keyword expansion in content strategy work.
+
+Call pattern: discovery → `get_report_schema` → `execute_report`.
+
+## Key reports
+
+- `organic_research` — organic keywords + top pages for a domain
+- `keyword_research` — expansion, volume, KD, intent
+- `overview_research` — domain-level traffic/visibility
+
+## Example
+
+```
+mcp__semrush__organic_research(query="<competitor.com>", database="us")
+mcp__semrush__keyword_research(query="<seed topic>", database="us")
+mcp__semrush__get_report_schema(report_id="<id>")
+mcp__semrush__execute_report(report_id="<id>", params={...})
+```
+
+Korean market: `database="kr"`.
+
+## Strengths for this skill
+
+- Best overall keyword + organic competitor view.
+- Cleanest discovery → schema → execute pattern for batch content brief generation.
+- Pairs with OurSEO crawl (on-page) and GSC (first-party performance).
+
+## Not for this skill when
+
+- **On-page inventory of target site** — OurSEO `crawl_website`.
+- **Cannibalization detection** — GSC first-party `combined` report.
+- **Korean Naver content** — `our research naver *` covers Naver-specific patterns Semrush doesn't.