[schemas] Full-text search — keyword search RPC for thoughts

[alanshurafa]

Summary

• Adds search_thoughts_text RPC using PostgreSQL's built-in full-text search
• Creates GIN index on thoughts content for fast keyword queries
• Supports natural language queries, phrase search, and boolean operators via websearch_to_tsquery
• Complements the existing match_thoughts semantic vector search

Why

Semantic search finds related concepts, but sometimes you need exact keyword matches — person names, project codes, dates, technical terms. This gives OB1 users both discovery (semantic) and retrieval (keyword) search modes.

Scale

Tested against 75,000+ thoughts in production. GIN index keeps queries fast.

Test plan

• Keyword search returns relevant results ranked by relevance
• Phrase search works with quoted strings
• Boolean operators (OR, AND, NOT) work
• GIN index prevents sequential scans on large tables
• RPC callable from Supabase client and Edge Functions

🤖 Generated with Claude Code

[justfinethanku]

Saw your Discord post — impressive work across all of these. I'm going to personally review each one. Not today, but they're on my list.

[justfinethanku]

Thank you for this contribution! I've completed a thorough review of PR #53. Here's my assessment:

What's Good

Full-Text Search (schemas/):

• Excellent documentation with clear use cases
• Clean SQL implementation using PostgreSQL's native full-text search
• Proper GIN indexing for performance
• Good complementary capability to existing semantic search
• Well-structured metadata.json with all required fields
• Solid troubleshooting section
• Production-tested at scale (75K+ thoughts)

Content Fingerprint Dedup (primitives/):

• Comprehensive README with excellent technical depth
• Strong "Extensions That Use This" section (required for primitives)
• Proper cross-reference to email-history-import recipe
• Clear step-by-step guide with batching strategies
• Good troubleshooting coverage

Issues to Address

Critical

1. PR Title Mismatch
Your PR title is "[schemas] Full-text search" but you're modifying TWO categories:

• primitives/content-fingerprint-dedup/
• schemas/full-text-search/

Resolution needed: Update the PR title to reflect both contributions. Suggested format:

[schemas][primitives] Full-text search RPC + content fingerprint dedup

2. PR Description Incomplete
The PR description only describes the full-text search feature. It doesn't mention that you're also adding the content fingerprint dedup primitive.

Resolution needed: Update the PR description to include:

• A summary of both contributions
• Why they're being submitted together (if there's a relationship)
• OR split into two separate PRs if they're independent

Moderate

3. Missing Temporal Metadata
The content-fingerprint-dedup/metadata.json is missing optional but recommended fields:

• created (should be "2026-03-16" based on the full-text-search dates)
• updated (should be "2026-03-16")

Resolution needed: Add these fields to match the pattern in schemas/full-text-search/metadata.json.

Questions for Clarification

4. Primitive Curation Process
Per CONTRIBUTING.md, primitives are CURATED and must be referenced by 2+ extensions. The README shows one reference (email-history-import).

Question: What is the second extension/recipe that uses this primitive? Or is this being submitted with maintainer pre-approval?

SQL Safety Check

✓ No DROP TABLE/DATABASE
✓ No TRUNCATE
✓ No unqualified DELETE FROM
✓ ALTER TABLE only adds columns (allowed per guard rails)
✓ No credentials or secrets

Automated Review Criteria

✓ Folder structure correct for categories
✓ Required files present (README.md + metadata.json for both)
✓ metadata.json valid and complete
✓ No binary blobs
✓ README sections complete
✓ Cross-references resolve (email-history-import exists)

Verdict: Minor fixes needed

This is quality work with solid technical implementation and good documentation. The issues are primarily administrative (title/description clarity) rather than technical problems. Once you address the title mismatch and PR description, this should be ready to merge.

Action items:

1. Update PR title to reflect both contributions
2. Update PR description to document both contributions
3. Add created and updated dates to content-fingerprint-dedup metadata.json
4. Clarify the second extension reference for the primitive (or confirm maintainer pre-approval)

Great contribution overall! The full-text search implementation is particularly clean and will be valuable for users who need exact keyword matching alongside semantic search.