geo-sleuth

What it does

• One-photo interface: hand the agent a photo and say “find where this was taken” — back comes camera position, heading and a satellite evidence image
• Works when there is nothing to read: OSM geometry, elevation skylines, satellite tiles and street view replace text and landmark clues
• Scripts rank, the model judges: ~20 single-purpose scripts search, score and sort; conclusions must cite the command run and the file produced
• Answers carry an error radius, graded confidence and overlay checks — distance rulers and skyline fits are shown, not asserted

When to use it

Geolocation-grade photo forensics from a single image — verifying where a claim photo was shot, tracing landscapes for research or journalism, or exercising an agent on a long multi-step evidence task where every step must point at a file.

Getting started

npx skills add Oldcircle/geo-sleuth
# pick your agents when prompted; or all six, user-wide:
npx skills add Oldcircle/geo-sleuth -g -a claude-code -a codex -a cursor -a gemini-cli -a opencode -a github-copilot -y

# requires Python 3.10+; then just:
#   find where this photo was taken

Things to know

Runs are long: the showcase case took about 72 minutes, roughly half of it computation. Rulers rely on stated assumptions (the case assumed a 32 m pier span), and accuracy degrades where OSM or elevation data is thin — the answer's error radius exists for a reason. Geolocation is dual-use: locating people or sensitive sites from photos can cause real harm, so use it on lawful targets. MIT.