Oldcircle/geo-sleuth

What it solves

geo-sleuth is an agent skill designed to pinpoint the exact location where a photo was taken, even when the image lacks GPS metadata, text, landmarks, or recognizable signs. It transforms visual clues—such as skyline shapes, pier spacing, and shadow angles—into precise geographic coordinates.

How it works

The project operates through a three-layer system where specialized Python scripts handle the heavy lifting of data processing and ranking, while the AI model acts as the final judge.

1. Decide: A candidate board (board.py) manages candidates, scores evidence, and determines the next search step.
2. Perceive: A suite of scripts performs specific tasks: intake.py handles EXIF and OCR; terrain.py matches ridge lines against elevation data; osm.py pulls geometry from OpenStreetMap; and others handle satellite and street view matching.
3. Judge: The AI model uses the SKILL.md instructions and references to propose hypotheses and select the most likely location from the ranked results.

Who it’s for

This tool is for users of AI agents (such as Claude Code, Cursor, GitHub Copilot, and Gemini CLI) who need to perform advanced geolocation (OSINT) tasks using a standardized "Agent Skill" format.

Highlights

• Geometry-Based Localization: Uses pier spacing as a distance ruler and ridge lines as fingerprints to match against elevation data.
• High Precision: Capable of locating photos to within a few meters in complex environments (e.g., rice paddies or city streets).
• Agent-Agnostic: Works with any agent that can read SKILL.md and execute shell commands.
• Evidence-Driven: Every conclusion must be backed by a specific command output and a produced file, avoiding guesswork based on population or fame.