Geospatial AI Skills for Claude

This directory contains markdown files that act as skills for Claude — structured context files you load into a conversation to give Claude deep, platform-specific knowledge about geospatial work on BigQuery or Snowflake.

Getting started

1. Clone this repository

git clone https://github.com/wriglz/geo-ai.git ~/geo-ai

Pull the latest rules at any time:

git -C ~/geo-ai pull

2. Point your project's CLAUDE.md at the relevant skill file

Rather than copying the file contents (which would go stale), use an @import directive so Claude always reads the latest version from your local clone:

Open or create a CLAUDE.md file in the root of your project and add one of the following lines:

<!-- BigQuery projects -->
@~/geo-ai/bigquery-spatial.md

<!-- Snowflake projects -->
@~/geo-ai/snowflake-spatial.md

<!-- Both platforms — import both -->
@~/geo-ai/bigquery-spatial.md
@~/geo-ai/snowflake-spatial.md

Claude Code resolves @-prefixed paths relative to your home directory and injects the file contents automatically at the start of every session in that project.

3. Using the skills in Claude.ai (web or desktop)

1. Open a new conversation with Claude.
2. Click the paperclip / attachment icon and upload the relevant .md file.
3. Claude will follow the rules in the file for the rest of the conversation.

You can upload multiple files if a pipeline touches both platforms.

Tips

• git pull before starting work on a new task to pick up any rule updates.
• Import only the file that matches your target platform — import both only if your project genuinely spans both platforms.
• Skills take effect immediately — you do not need to repeat the rules in later messages.

---

Files

bigquery-spatial.md

BigQuery skill. Covers clustering with S2 geography columns, geometry ingestion with make_valid, spatial index columns (H3 requires CARTO Toolbox), query patterns that benefit from clustering, and common anti-patterns.

snowflake-spatial.md

Snowflake skill. Covers clustering with ST_GEOHASH, the GEOMETRY vs GEOGRAPHY distinction, safe constructors (TRY_TO_GEOGRAPHY), predicates that disable micro-partition pruning, memory spillage on spatial operations, and GeoJSON properties in VARIANT columns.