Fix video list pagination cursors not encoded with provider metadata

[tsachis]

Summary

• transform_video_list_response: first_id and last_id pagination cursor fields are now encoded with encode_video_id_with_provider(), matching the encoding already applied to data[].id
• transform_video_list_request: the after parameter is now decoded via extract_original_video_id() before being sent upstream, so wrapped cursor IDs round-trip correctly
• Added 6 unit tests covering encoding, decoding, passthrough, missing fields, and full round-trip pagination

Fixes #20708

Root Cause

In litellm/llms/openai/videos/transformation.py, transform_video_list_response only iterated over data[] to encode video IDs with provider metadata, but left first_id and last_id as raw provider IDs. Clients using these cursors for pagination would pass unencoded IDs back, breaking provider routing.

Test plan

• TestVideoListTransformation::test_transform_video_list_response_encodes_first_id_and_last_id — verifies both cursors are encoded
• TestVideoListTransformation::test_transform_video_list_response_no_provider_leaves_ids_unchanged — no encoding when provider is None
• TestVideoListTransformation::test_transform_video_list_response_missing_pagination_fields — handles absent first_id/last_id
• TestVideoListTransformation::test_transform_video_list_request_decodes_after_parameter — encoded after cursor decoded correctly
• TestVideoListTransformation::test_transform_video_list_request_passes_through_plain_after — plain IDs pass through
• TestVideoListTransformation::test_transform_video_list_roundtrip — full round-trip: list → use last_id as after → correctly decoded
• All existing video generation and Azure video transformation tests pass

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 8, 2026 2:26pm

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Overview

Greptile Summary

This PR correctly fixes the pagination cursor encoding issue in video list operations. The first_id and last_id fields are now properly encoded with provider metadata to match the encoding already applied to data[].id, and the after parameter is decoded on incoming requests.

Key changes:

• transform_video_list_response: Encodes first_id and last_id with provider metadata using encode_video_id_with_provider(), extracting the model from the corresponding first/last item in the data array
• transform_video_list_request: Decodes the after parameter using extract_original_video_id() to unwrap encoded cursor IDs before sending upstream
• Added 6 comprehensive unit tests covering encoding, decoding, passthrough behavior, missing fields, and full round-trip pagination

The implementation follows the existing pattern used for data[].id encoding and correctly handles edge cases like missing pagination fields and plain (non-encoded) IDs.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are well-isolated to video list pagination logic, follow the existing encoding pattern already used for video IDs, include comprehensive test coverage (6 unit tests covering all edge cases), and fix a legitimate bug where pagination cursors were inconsistent with the encoded video IDs
• No files require special attention

Important Files Changed

Filename	Overview
litellm/llms/openai/videos/transformation.py	Adds encoding to first_id and last_id pagination cursors, and decodes after parameter in requests - logic is correct and consistent with existing ID encoding pattern
tests/test_litellm/test_video_generation.py	Adds 6 comprehensive unit tests covering encoding/decoding, edge cases, and full round-trip pagination scenarios - excellent test coverage

Sequence Diagram

sequenceDiagram
    participant Client
    participant LiteLLM
    participant OpenAI as OpenAI/Azure Provider

    Note over Client,OpenAI: Video List - First Page Request
    Client->>LiteLLM: GET /v1/videos?limit=10
    LiteLLM->>LiteLLM: transform_video_list_request(after=None)
    LiteLLM->>OpenAI: GET /v1/videos?limit=10
    OpenAI-->>LiteLLM: {data: [{id: "video_aaa"}, {id: "video_bbb"}],<br/>first_id: "video_aaa", last_id: "video_bbb"}
    LiteLLM->>LiteLLM: transform_video_list_response()
    LiteLLM->>LiteLLM: encode data[].id with provider metadata
    LiteLLM->>LiteLLM: encode first_id with provider metadata (NEW)
    LiteLLM->>LiteLLM: encode last_id with provider metadata (NEW)
    LiteLLM-->>Client: {data: [{id: "video_encoded_aaa"}],<br/>first_id: "video_encoded_aaa",<br/>last_id: "video_encoded_bbb"}

    Note over Client,OpenAI: Video List - Next Page Request
    Client->>LiteLLM: GET /v1/videos?after=video_encoded_bbb
    LiteLLM->>LiteLLM: transform_video_list_request(after="video_encoded_bbb")
    LiteLLM->>LiteLLM: extract_original_video_id() → "video_bbb" (NEW)
    LiteLLM->>OpenAI: GET /v1/videos?after=video_bbb
    OpenAI-->>LiteLLM: {data: [{id: "video_ccc"}],<br/>first_id: "video_ccc", last_id: "video_ccc"}
    LiteLLM->>LiteLLM: transform_video_list_response()
    LiteLLM->>LiteLLM: encode pagination cursors with provider metadata
    LiteLLM-->>Client: {data: [{id: "video_encoded_ccc"}],<br/>first_id: "video_encoded_ccc",<br/>last_id: "video_encoded_ccc"}

Loading

[greptile-apps]

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}