[Lucas Kim (kimnamu)]

Motivation & Context

Thanks for the Bedrock connector and for the Anthropic client's image mapping, which this PR mirrors. User-message images were skipped before Converse, so GPT-6 Sol/Luna/Astra answered blind.

Description & Review Guide

• What are the major changes? A case "data" if content.has_top_level_media_type("image") branch returns a Converse image block via the existing _get_data_bytes (as Gemini does). Images outside user messages stay skipped. Tests cover png/jpeg (fail without the fix) and an assistant image.
• What is the impact of these changes?

Red 64x64 PNG, "What color is this image? One word.", live us-east-1	Before	After
us.openai.gpt-6-sol	❌ 'Unknown.', blocks sent ['text']	✅ 'Red', blocks sent ['text', 'image']
us.openai.gpt-6-luna	❌ 'Gray', blocks sent ['text']	✅ 'Red', blocks sent ['text', 'image']
us.openai.gpt-5.6-sol	❌ 'Unknown'	✅ 'Red'
us.openai.gpt-6-astra	❌ 'Unknown.', blocks sent ['text']	✅ 'Red.', blocks sent ['text', 'image']
global.openai.gpt-6-astra	❌ 'Unknown', blocks sent ['text']	✅ 'Red', blocks sent ['text', 'image']
Image in an assistant message (history)	skipped, request OK	✅ unchanged (sending it gets a ValidationException from GPT-6 Sol/Luna/Astra)
Non-image data (application/octet-stream, text/plain)	dropped	✅ still dropped, existing tests unchanged
Rich content in tool results	dropped with a warning	✅ unchanged
Public API	✅	✅ unchanged

$ python live.py   # before
us.openai.gpt-6-sol -> 'Unknown.' | user blocks sent ['text']
us.openai.gpt-6-luna -> 'Gray' | user blocks sent ['text']
$ python live.py   # after
us.openai.gpt-6-sol -> 'Red' | user blocks sent ['text', 'image']
us.openai.gpt-6-luna -> 'Red' | user blocks sent ['text', 'image']
$ python live.py us.openai.gpt-6-astra global.openai.gpt-6-astra   # before
us.openai.gpt-6-astra -> 'Unknown.' | user blocks sent ['text']
global.openai.gpt-6-astra -> 'Unknown' | user blocks sent ['text']
$ python live.py us.openai.gpt-6-astra global.openai.gpt-6-astra   # after
us.openai.gpt-6-astra -> 'Red.' | user blocks sent ['text', 'image']
global.openai.gpt-6-astra -> 'Red' | user blocks sent ['text', 'image']

• What do you want reviewers to focus on? Format = lower-cased media subtype, with jpg mapped to jpeg. Subtypes Converse doesn't accept (its enum is [gif, jpeg, png, webp]), such as image/bmp or image/svg+xml, are skipped with a warning like other content this client can't send, so the text still goes through.

Related Issue

Fixes #8682

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.

[Copilot]

Copilot review overview

🟡 Changes recommended

Image formats require validation, and the private helper import needs its required Pyright suppression.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 2

Open (2)

• Add suppression for private helper import · New
• Whitelist supported image formats and narrow optional media type · New

What changed in this PR

Adds Bedrock Converse support for images in user messages while preserving existing behavior for assistant images.

Changes:

• Converts PNG/JPEG user image data into Converse image blocks.
• Adds regression coverage for user and assistant image handling.

File	Description
python/​packages/​bedrock/​agent_framework_bedrock/​_chat_client.py	Adds Bedrock image-block conversion and role filtering.
python/​packages/​bedrock/​tests/​test_bedrock_client.py	Tests image conversion and assistant-image skipping.

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

[Eduard van Valkenburg (eavanvalkenburg)]

Lucas Kim (@kimnamu) The current head is green, and the dedicated full-diff review found no remaining actionable issues with Bedrock user-image forwarding, user-role filtering, supported-format validation, unsupported-format handling, or the accompanying regression coverage.