The vendor lane (--use-official-api) caught this during the 1.17 gate: Anthropic's Files listing no longer speaks the cursor scheme the gateway serves.
What upstream does now
GET /v1/files returns data and next_page, an opaque token. It no longer returns first_id, last_id or has_more, and the SDK (anthropic 1.0.0, 2026-08-20) takes a page parameter instead of after_id / before_id.
The change is visible in anthropic.pagination.SyncPageCursor, which declares exactly data and next_page; the three fields the gateway sends arrive only as model_extra, and against the official API they are absent altogether.
What the gateway does
stdapi/files/_core.py still serves the ID-cursor envelope: first_id, last_id, has_more, and after_id / before_id as query parameters. That is what its own clients were written against, and it keeps working — but it is now a divergence from upstream rather than a match.
Marked, not hidden
Three tests in tests/test_anthropic_files.py record the split rather than papering over it:
test_anthropic_list_after_id and test_anthropic_list_before_id carry @pytest.mark.gateway(...), so they no longer claim upstream parity.test_list_files stops before the edge-ID assertions on the official lane.
Decide
Whether to adopt page / next_page — as a replacement, or alongside the ID cursors for the clients already using them — is a compatibility decision, not a bug fix, so it is deliberately not in 1.17.
The vendor lane (
--use-official-api) caught this during the 1.17 gate: Anthropic's Files listing no longer speaks the cursor scheme the gateway serves.What upstream does now
GET /v1/filesreturnsdataandnext_page, an opaque token. It no longer returnsfirst_id,last_idorhas_more, and the SDK (anthropic1.0.0, 2026-08-20) takes apageparameter instead ofafter_id/before_id.The change is visible in
anthropic.pagination.SyncPageCursor, which declares exactlydataandnext_page; the three fields the gateway sends arrive only asmodel_extra, and against the official API they are absent altogether.What the gateway does
stdapi/files/_core.pystill serves the ID-cursor envelope:first_id,last_id,has_more, andafter_id/before_idas query parameters. That is what its own clients were written against, and it keeps working — but it is now a divergence from upstream rather than a match.Marked, not hidden
Three tests in
tests/test_anthropic_files.pyrecord the split rather than papering over it:test_anthropic_list_after_idandtest_anthropic_list_before_idcarry@pytest.mark.gateway(...), so they no longer claim upstream parity.test_list_filesstops before the edge-ID assertions on the official lane.Decide
Whether to adopt
page/next_page— as a replacement, or alongside the ID cursors for the clients already using them — is a compatibility decision, not a bug fix, so it is deliberately not in 1.17.