Geolocation OSINT skills for AI agents

Activate this skill when the user is using the built environment in a photograph or video to narrow a location: utility poles and insulators, power-line styles, road surfaces, kerb design, house styles and roofs, window shapes and shutters, fire hydrants, manhole covers, postboxes, street furniture and chain stores. Triggers on "utility pole clue," "what country are these buildings," "roof style geolocation," "manhole cover," "fire hydrant type," "chain store by country," "kerb painting," or "infrastructure clues." Covers what each object says, the national systems that make it say it, and how to combine several weak infrastructure clues into a strong regional fix.

Activate this skill when the user wants to work out where a photograph or video frame was taken by reading the scene itself: sky, horizon, terrain, roads, signage, vehicles, people and text. Triggers on "geospatial clue," "geolocation," "image geolocation," "where was this taken," "geolocate this photo," "clue log," "hypothesis ranking," or "verify the location of this video." Covers the systematic zone-by-zone pass over an image, how to record every observation separately from its interpretation, and how to form, rank and try to falsify competing location hypotheses before touching a map.

Activate this skill when the user needs to establish where an image or video came from before or alongside geolocation: reading EXIF, XMP and container metadata and interpreting its absence, reverse image search, finding the earliest appearance online, detecting edits and recompression, extracting frames from video, and keeping chain-of-custody notes. Triggers on "EXIF data," "metadata stripped," "reverse image search," "earliest upload," "is this photo edited," "extract video frames," "ffprobe," "exiftool," "chain of custody," or "provenance." Covers the commands, the traces each platform leaves, and the evidence log that makes a finding defensible.

Activate this skill when the user is narrowing an image geolocation using road furniture: sign shapes and colours, line markings, chevrons, bollards and delineator posts, guardrails, kilometre posts and the side of the road traffic drives on. Triggers on "road signs by country," "centre line colour," "chevron signs," "bollards," "delineator posts," "guardrail," "kilometre marker," "driving side," "Vienna Convention signs," or "road furniture reference set." Covers reading each feature, the national systems behind them, and how to build and maintain a verified reference set instead of relying on memory.

Activate this skill when the user has narrowed an image geolocation to a region and needs to find the exact spot: turning clues into a search area, querying map data for candidate features, matching roof shapes and road geometry from above, using historical imagery to bound dates, confirming with street-level imagery and documenting the match. Triggers on "find this on satellite," "match the roofs," "Overpass query," "street view confirmation," "historical imagery," "camera position," "document the geolocation," or "how to prove the match." Covers the imagery sources and their blind spots, the geometry of turning a photo into a plan view, and the evidence standard for a confirmed fix.

Activate this skill when the user wants to extract time of day, date, hemisphere, latitude or camera bearing from the sun and the shadows in a photograph or video. Triggers on "shadow analysis," "sun position," "solar azimuth," "what time was this taken," "shadow length," "chronolocation," "SunCalc," "hemisphere from the sun," or "combine shadows with EXIF." Covers turning shadow direction and length into bearing and time, the solar geometry behind it, calculator tools, combining the result with metadata and map bearings, and stating honest error bars.

Activate this skill when the user is using the natural environment in an image to constrain where it was taken: biomes and indicator plants, soil colour, snow lines and treelines, coastline shapes and mountain profiles, matched against elevation data and climate maps. Triggers on "what climate is this," "identify the terrain," "mountain skyline match," "biome from photo," "soil colour clue," "snow line," "coastline shape," "Köppen zone," "DEM match," or "vegetation geolocation." Covers reading the landscape, the reference datasets that describe it, and the procedure for turning a skyline into a bearing and a search area.

Activate this skill when the user is extracting location evidence from vehicles in a photograph or video: number plate formats and colours by country and region, car models by market, taxis and buses as regional markers, fleet and emergency-service liveries, and the privacy rules for handling what plates reveal. Triggers on "license plate format," "number plate colour," "which country is this plate," "taxi colours," "bus livery," "car models by country," "plate region code," "police car livery," or "blur the plates." Covers reading a plate even when it is partly obscured, regional codes that turn a plate into a district, and the handling of personal data.

Activate this skill when the user must decide how confident a geolocation is, how to get it independently confirmed, and whether and how to publish it: confidence levels, second-analyst review, avoiding doxxing, withholding locations that endanger people, minimising harm to those visible in images, and writing up the methodology. Triggers on "confidence level," "verification standard," "second analyst," "should we publish the location," "doxxing risk," "blur faces," "duty of care," "methodology write-up," or "verification ethics." Covers the standard a location finding must meet before it is reported, the editorial and human-rights considerations that can override publication, and the write-up that lets others check the work.