feat: Linear and Notion live-sync recipes + fix embedding config key lookup

[mishanefedov]

What this adds

Two new integration recipes following the existing deterministic-collector pattern (code for data, LLMs for judgment), plus a bug fix.

recipes/linear-to-brain.md

Live sync of Linear issues, projects, and cycles into brain pages for offline semantic search. Complements MCP real-time access — gbrain query works without an active session or network connection.

• GraphQL-based collector with full pagination (pageInfo.hasNextPage + endCursor)
• Incremental sync via updatedAt filter (only changed issues re-fetched)
• Priority mapping (0–4 → human labels) for searchable frontmatter
• Month-partitioned output: brain/linear/issues/{YYYY-MM}/{id}.md
• Single credential: LINEAR_API_KEY (read-only, no OAuth)
• Daily launchd/cron sync recipe included

recipes/notion-to-brain.md

Live Notion API sync as a complement to the existing one-shot migrate skill. Handles the ongoing case: new pages appear in the brain automatically on a schedule.

• Search-based discovery (POST /v1/search) — no hardcoded page IDs
• Full block-to-markdown renderer (paragraph, headings, lists, to_do, code, divider, child_page)
• Database row support: properties rendered as typed frontmatter
• Incremental sync via last_edited_time
• 350ms request throttle to respect Notion's 3 req/s limit
• Documents the sharing-required constraint (Notion API security model)

skills/RESOLVER.md

Added routing entries for both recipes, disambiguating the new live-sync paths from the existing one-shot Notion migration skill.

src/core/embedding.ts (bug fix)

new OpenAI() reads only from OPENAI_API_KEY env var. If the key is stored in ~/.gbrain/config.json (the openai_api_key field that loadConfig() already defines), the embedding step silently fails. Fix: fall back to loadConfig()?.openai_api_key when the env var is absent. Env var still takes precedence.

Tests

bun test → 1191 pass, 0 fail (including test/integrations.test.ts which validates recipe frontmatter schema)

E2E tests skipped (no DATABASE_URL) — new recipes don't touch the engine layer, covered by the integration frontmatter tests.

Fits the welcome PRs list

PRs welcome for: new enrichment API integrations

Both recipes fall squarely in this category.