Objects in a bucket, paged (?prefix=&delimiter=&after=&limit=)

What is actually IN one of your org's storage buckets - the listing an agent needs before importing anything. list_connections names your buckets (kind "gmn-storage", shown as storage/<bucket>); this shows their contents. Pass delimiter: "/" to walk the bucket a directory at a time: every key sharing the next segment folds into prefixes (e.g. ["train/", "val/"]) instead of listing its members, so a bucket of a million objects browses in a handful of calls - then re-call with that prefix to descend. Omit delimiter for a flat list of every key under prefix. Returns {bucket, prefix, delimiter, objects: [{key, size_bytes, created_at}], prefixes, truncated, next_after}, keys in byte order (the order every S3 client's sync assumes). truncated: true means there is another page: call again with after: <next_after>, unchanged prefix and delimiter. limit bounds one call (1-1000, default 1000; keys and folded prefixes count together). This reads the object ledger, never the bytes: no node is touched, nothing is downloaded, no egress is billed. To land bytes on a node, import_data(connection: "storage/<bucket>", source: "<key>" - the key exactly as it reads here, bucket-relative, or the full "s3://<bucket>/<key>"; a "/"-terminated prefix for the subtree, or a non-slash key prefix to take a flat prefix in chunks); a human's own terminal has the same listing as autor storage ls.