[sdairs]

Query API endpoint lists return an array in result, with cursor metadata on the response envelope. The client decoded a nested items/pagination object instead, causing the scheduled Cloud Integration lifecycle and CLI list/name lookups to fail with invalid type: map, expected a sequence. A nonempty array is interpreted as positional struct fields; {"result":[{}]} reproduces the exact CI error at line 1, column 11.

Return ApiResponse<Vec<PublicQueryApiEndpointListItem>> from the library and consume its envelope pagination in the lifecycle test. The CLI adapts the response to its existing items/pagination output. Existing command flags, requests, authentication and name-selection safety checks are preserved. The legacy nested container remains exported for callers that use it independently; the library README documents the return-type migration.

The corrected client/model and subprocess fixtures cover nonempty and empty arrays, sparse/null responses, envelope metadata, opaque cursors, OAuth, multiple pages and unsafe name-selection failures.

Stacked directly above #1094 (claude/drift-cli-1088). Related release acceptance: #757; this PR does not close that tracker.

This is separate from #1088: neither its current gap list nor the normal dry-run at #1094 identifies the endpoint-list mismatch. The dry-run still reports the same 27 remaining findings after this change; no snapshot or analyzer policy is changed.

Validation:

• The corrected client regression failed with the old decoder and passes with this fix.
• Passed formatting, both CLI Clippy configurations, telemetry-disabled compilation, API/analyzer all-target Clippy and tests, and 96 Python/classifier tests.
• Full CLI tests: all targets passed except failed_fresh_start_preserves_postgres_identity_without_polluting_clickhouse_selection, which reports docker_unavailable because the local Docker socket is absent. Endpoint subprocess and pagination/name-selection tests passed.
• Read-only live smoke: the installed client reproduced the original decoding failure; the patched binary read two cursor pages and resolved an endpoint by name. No SQL ran and no Cloud resources changed.
• The full provisioning lifecycle was not rerun.

---

Note

Medium Risk
Breaking return type for Rust callers of query_api_endpoint_list, but behavior matches the live API and fixes list/name-resolution failures; CLI output shape is preserved via an adapter layer.

Overview
Fixes Query API endpoint list decoding so the Cloud API client matches the real wire shape: result is an endpoint array and pagination lives on the envelope (next_cursor, limit, total_count), not a nested items/pagination object. That mismatch caused JSON decode failures (e.g. expected a sequence) for nonempty lists in integration and CLI name lookups.

query_api_endpoint_list now returns ApiResponse<Vec<PublicQueryApiEndpointListItem>>; QueryApiEndpointListResponse stays exported as a legacy helper type. clickhousectl rebuilds items/pagination from the envelope so human/JSON list output and --cursor behavior stay the same; missing/null result now fails the command instead of printing an empty list.

Docs note envelope vs CLI pagination fields; client, model, integration, and CLI subprocess tests use the corrected list envelopes and multi-page cursors.

^{Reviewed by Cursor Bugbot for commit b84d7b9. Bugbot is set up for automated code reviews on this repo. Configure here.}