# EnriVision Changelog
All notable changes to EnriVision are documented in this file.
## 2026-01-13
### Tests
- Added a regression test to ensure EnriVision strips internal extraction fields (e.g., `multipass`, `model`, and upload/routing identifiers) before returning tool output.
## 2026-01-12
### Docs
- Documented Claude Code CLI `Read(...)` supported formats as a compatibility reference, clarifying when EnriVision MCP is required (unsupported images, video/audio, and larger/binary inputs).
## 2026-01-08
### Changed
- EnriProxy `/v1/vision/analyze` responses no longer include internal provider model identifiers or multipass routing details; EnriVision tool output remains safe and minimal.
- Added `extraction.timeline` for audio/video analyses and kept optional `extraction.segment_summaries` for multipass timelines.
- Raised EnriVision Node.js engine requirement to `>= 24` (aligned with the Enri monorepo standard).
### Improved
- Animated image key frame selection now prioritizes visually distinct frames (scene-change heuristic) for GIF/WebP/APNG.
## 2026-01-07
### Added
- Animated SVG/APNG uploads now produce representative key frames via EnriProxy for analysis.
- Audio uploads now recognize additional common extensions (AIFF/CAF/M4B/M4R/OGA/WEBA/MKA/MP1/MP2/MPA/MPGA).
- Added video clip window parameters (`video.clip_start_seconds`, `video.clip_duration_seconds`) for time-targeted analysis.
### Changed
- Updated npm publishing references to use the @bedolla/enrivision scope and refreshed README examples.
- Removed per-call server_url/api_key/timeout_ms overrides from MCP schema, tool parsing, and docs.
- Clarified animated GIF/WebP support in EnriVision tool descriptions and README.
- Refined README wording around upload size defaults and PDF truncation risk.
### Tests
- Added request payload tests for EnriVision's EnriProxy client (upload session + analyze payloads).
