Description
A friend sends a photo with no location data and asks where it was taken. There is no sign, no plate, no landmark, just a bridge and a mountain. Guessing gets you nowhere.
geo-sleuth is an image geolocation skill for AI coding agents. Install it, hand your agent a photo and say "find where this photo was taken", and it runs the scripts itself and comes back with the camera position, the direction it was facing, a satellite evidence image and coordinates with an error radius. It doesn't read text, it uses geometry: pier spacing becomes a distance ruler, shadows give a bearing, and a ridge line becomes a fingerprint matched against elevation data, alongside OpenStreetMap geometry, satellite imagery and street view to narrow the search step by step.
In the author's cases, a text-free photo of a rice paddy and a viaduct was narrowed from tens of thousands of rail-bridge segments to a final position within about 2 m. Every conclusion has to name the command that ran in the session and the file it produced, and population or fame doesn't count as evidence. Scripts search, score and rank, and the model only judges among the top few.
One sentence to a location: give your agent a photo and get the camera position, heading, a satellite evidence image and a graded confidence.
Works with nothing to read: OpenStreetMap geometry, elevation skylines, satellite tiles and street view carry the search without signs or landmarks.
Twenty single-purpose scripts: they search, score and sort, and each step leaves a file you can check.
Many agents: a standard Agent Skill (a SKILL.md plus plain Python scripts) that runs in Claude Code, Codex, Cursor, Gemini CLI, OpenCode and GitHub Copilot.
Easy install:
geo-sleuth is an image geolocation skill for AI coding agents. Install it, hand your agent a photo and say "find where this photo was taken", and it runs the scripts itself and comes back with the camera position, the direction it was facing, a satellite evidence image and coordinates with an error radius. It doesn't read text, it uses geometry: pier spacing becomes a distance ruler, shadows give a bearing, and a ridge line becomes a fingerprint matched against elevation data, alongside OpenStreetMap geometry, satellite imagery and street view to narrow the search step by step.
In the author's cases, a text-free photo of a rice paddy and a viaduct was narrowed from tens of thousands of rail-bridge segments to a final position within about 2 m. Every conclusion has to name the command that ran in the session and the file it produced, and population or fame doesn't count as evidence. Scripts search, score and rank, and the model only judges among the top few.
Features
One sentence to a location: give your agent a photo and get the camera position, heading, a satellite evidence image and a graded confidence.
Works with nothing to read: OpenStreetMap geometry, elevation skylines, satellite tiles and street view carry the search without signs or landmarks.
Twenty single-purpose scripts: they search, score and sort, and each step leaves a file you can check.
Many agents: a standard Agent Skill (a SKILL.md plus plain Python scripts) that runs in Claude Code, Codex, Cursor, Gemini CLI, OpenCode and GitHub Copilot.
Easy install:
npx skills add Oldcircle/geo-sleuth, then have the agent run doctor.py on first use. Needs Python 3.10 or newer, MIT licensed.
