01 · Product name
geo-sleuth
Free MIT skill that ranks a photo's place.
Published by Oldcircle.
In short
geo-sleuth is an agent skill in Oldcircle/geo-sleuth. The README says it finds where a photo was taken and shows its work. On 17 Oct 2026 the GitHub API reported 3,165 stars and 421 forks, the license field was MIT, and there was no release. It asks for Python 3.10 or newer and a SKILL.md install. The README names Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and GitHub Copilot. The first run is doctor.py. Scripts rank and the model judges. A place candidate is not a person's name.
Product pricing strategy
02
Free
geo-sleuth does not publish a price because it does not sell a seat. The GitHub license field on 17 Oct 2026 was MIT, there was no GitHub release, and there is no homepage with a checkout. You install the skill into an agent you already have. Those agents may have their own subscriptions. This listing does not price Claude, Codex, Cursor, Gemini, OpenCode, or Copilot. The project does not host a photo-search website for you. Satellite and street-level imagery the scripts consult may belong to other services. The README does not put a geo-sleuth fee on those calls, and it does not say they are free of every third-party limit.
Utility and features
03
The geo-sleuth README is the product page. There is no separate homepage. The repository was created on 18 Sep 2026, the default branch is main, and the last push recorded by the API on 17 Oct 2026 was 16 Oct 2026. The description says the skill uses OpenStreetMap geometry, elevation skylines, satellite imagery, and street view, and that it returns a camera position, a facing direction, and a satellite evidence image. No API-key requirement appears in the README sections checked for this listing. The interface the README describes is one sentence to the agent: find where this photo was taken. On first use it says to run doctor.py from the skill's scripts folder and fix any failed checks.
- An agent skill, MIT. SKILL.md, Python 3.10 or newer, no GitHub release on 17 Oct 2026. The license field was MIT.
- Show the file. The README says a conclusion has to name the command that ran and the file it produced.
- Scripts rank. Twenty scripts search and score. The model picks among the top few. Fame is not evidence.
The same README says a conclusion has to name the command that ran and the file it produced. Population and fame are not evidence. Twenty single-purpose scripts search, score, and sort, and the model only picks among the top few. That is a ranking of places, not an identity. This listing does not turn the README's worked landscape into steps for finding a person. A photo you are not allowed to examine is out of scope. The README is a badge row plus one worked example, which is not two product frames, so Screenshots are omitted.
- 1
Install the skill into one of the agents the README names, then run doctor.py before the first photo.
- 2
Ask for a place only on a photo you are allowed to examine. Keep the person out of the question.
- 3
Require the answer to name the command and the output file. Drop a rank that does not.
- 4
Write the result as a place candidate. Do not convert it into a person's identity.
Caveats
04
A ranked coordinate is a candidate for where a photo was taken. The README requires the conclusion to name the command and the file. If the note does not name them, the rank is unfinished. Population and fame are not evidence, which is the README's own limit. A famous skyline can still be the wrong city, and a model preference is not a measurement.
This listing does not explain how to locate a private person from a picture. A place is not a name, and a facing direction is not an identity. The skill is not a live camera and not a device-location feed. Star counts move. 3,165 was the API figure on 17 Oct 2026. There is no pinned release to cite, so cite the commit you installed. One worked example in the README is not a validation set.
Potential use cases
06
- A photo of a public place needs a location hypothesis. geo-sleuth is the MIT skill the README describes: Python 3.10 or newer, a SKILL.md file, and an agent the README names, such as Claude Code, Codex, Cursor, Gemini CLI, OpenCode, or GitHub Copilot. On first use the README says to run doctor.py and fix failed checks. There was no GitHub release on 17 Oct 2026, so record the commit you installed. The API star count that day was 3,165.
- The agent returns a camera position, a facing direction, and a satellite evidence image. The README says the conclusion has to name the command that ran and the file it produced, and that population and fame are not evidence. Scripts rank and the model picks among the top few. Attach the file. If the note only says the model was confident, it does not meet the README's bar. A famous skyline can still be the wrong place.
- Someone asks for the person who took the photo, or for where a private individual lives. Stop. The skill ranks a place for a photo you are allowed to examine. A coordinate is not a name, and this listing will not add steps that turn a picture into a person. Raven and Copernicus Browser answer different questions, one as a qualified image-geolocation platform and one as a Sentinel viewer. Do not mix their outputs into a single identity.
Hard cases
The case where the first result is easy to over-read.
A landmark rank is written as the photographer
The README's ranking can prefer a well-known view. That preference is not a measurement, which is why the project says fame is not evidence and why the command and the file have to be named. Write the candidate place, the file, and the checks you did not run. Do not add a person's name. There is no separate geo-sleuth website to cite, and there is no pinned release, so the repository commit is the version.
Practitioners using it
Posts on X from people describing a concrete use. The quote is the post. Likes and reposts are the counts saved with it, not a live total. A label marks a creator, a maintainer, a data partner, or an investor.
รูปเตาอบริมนา ไกลๆ มีสะพานรถไฟกับภูเขา ไม่มีพิกัดในไฟล์ ไม่มีตัวหนังสือสักตัว geo-sleuth คือสกิลที่ให้ AI agent สืบว่ารูปนี้ถ่ายที่ไหน ไล่ตัดสะพานรถไฟ 27,335 ช่วงจนเหลือจุดเดียว เคสที่ Oldcircle ผู้พัฒนาโชว์ไว้ พลาดไม่เกิน 2 เมตร แต่ตัวเต็งอันดับ 1 ตอนแรกกลับเป็นเมืองผิด 1/12 pic.twitter.com/QPz95QAQwU
I tested geo-sleuth, an AI agent skill that can find where a photo was taken and give you the coordinates. And honestly, it's awesome! I tested it with random photos in Claude Code. It correctly identified the locations in 3 examples, but failed on another. How to try it:… pic.twitter.com/ZDS6iRJSYH
离谱！这不是抖音很多博主都在做吗？ 一张没有 GPS、没有路牌的照片，AI 能告诉你在哪拍的 geo-sleuth：27 天 2800 星。给 Agent 一张照片，它靠桥墩间距、影子方向、山脊线，算出拍摄位置，还附卫星图当证据。README 里一个案例误差只有 2 米。 下次朋友发"猜猜我在哪"，你直接甩给它......…
Finding where a photo was taken shouldn’t stop at a guess. geo-sleuth is an agent skill for investigating where a photo was taken, including images with no readable text, plates, or obvious landmarks. It helps you move from a photo to a documented location hypothesis by… pic.twitter.com/b93SeF79cj
You can use Claude Code to find where a street photo was taken, within three meters, in under an hour. A free skill called geo-sleuth did it on a Vancouver cherry tree photo with no GPS data. Get it here: https://t.co/xIXwiL17Xn pic.twitter.com/Aly6lruXTs
Questions
Is geo-sleuth free?
The repository is MIT and had no release or paid tier on 17 Oct 2026. The agent you run it inside may be a separate subscription.
Does a geo-sleuth rank identify a person?
No. The README ranks a camera position and requires the command and the file. A place candidate is not a name.
Citations
07
Titled sources. These public pages are what the quotes on this listing come from.
Product
geo-sleuth repository
https://github.com/Oldcircle/geo-sleuth