Notion skill: add Keystone MCP fallback and large page handling

When mcp__notion__notion-fetch isn't available (e.g., Thanx/Keystone
environments), the skill now documents fallback tools for reading pages,
searching, and querying databases. Also adds guidance for handling large
page responses that exceed token limits.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: drn

agents/skills/notion/SKILL.md

@@ -14,6 +14,20 @@ Read Notion pages and databases via MCP tools. This skill covers reading operati
 | `mcp__notion__notion-fetch` | Read page content as Notion-flavored markdown |
 | `ReadMcpResourceTool` with `notion://docs/enhanced-markdown-spec` | Get the full markdown spec (useful for understanding page structure) |
 
+### Fallback: Keystone MCP Tools
+
+If `mcp__notion__notion-fetch` is not available (e.g., in environments using Keystone), use these tools instead:
+
+| Tool | Use For |
+|------|---------|
+| `mcp__plugin_thanx_keystone__notion_get_page` | Read page metadata (title, properties, dates) |
+| `mcp__plugin_thanx_keystone__notion_get_page_content` | Read page blocks (raw Notion API JSON) |
+| `mcp__plugin_thanx_keystone__notion_search` | Search for pages by title |
+| `mcp__plugin_thanx_keystone__notion_query_database` | Query a database with filters |
+| `mcp__plugin_thanx_keystone__notion_get_database` | Get database schema and metadata |
+
+The blocks API returns raw JSON. To extract readable text, parse each block's `rich_text[].plain_text` fields. Use ToolSearch to discover and load these tools before calling them.
+
 ## Instructions
 
 You are helping read Notion data. Use the MCP tools above for all operations.
@@ -53,7 +67,15 @@ These indicate sections that were absorbed into code blocks during a bad edit.
 - **Page reading**: Full page content as structured markdown
 - **Database browsing**: View database schemas and entries
 
+### Large Page Handling
+
+Notion pages can return very large responses (100K+ characters) that exceed token limits. When this happens:
+- The content may be saved to a temp file automatically
+- Use `python3 -c` or `jq` to extract block text rather than reading raw JSON directly
+- For Keystone blocks JSON, extract text with: `jq -r '.. | .rich_text? // empty | .[].plain_text' < temp_file.json`
+- Consider requesting specific sections rather than the full page when possible
+
 ## Limitations
 
 - **Read-only** — Cannot create, update, or delete pages
-- Page content is returned as Notion-flavored markdown, not raw JSON
+- Page content is returned as Notion-flavored markdown, not raw JSON (except Keystone fallback which returns raw blocks)