feat: add ImageBlock vision support for multimodal chat

[robotdad]

Summary

• Implements ImageBlock vision support in Anthropic provider, enabling multimodal chat with images
• Adds conversion from Amplifier's ImageBlock protocol format to Anthropic's vision message format
• This is the second provider (after Gemini) to implement image support, validating the protocol design

Implementation Details

The implementation handles the conversion from Amplifier's protocol ImageBlock format to Anthropic's vision API format in _convert_messages:

• Converts ImageBlock objects with URL/base64 data and media type
• Properly structures content as arrays when images are present
• Maintains backward compatibility with text-only messages

API Format Differences

Anthropic uses a different vision format compared to Gemini:

• Gemini: inline_data with base64
• Anthropic: image blocks with source object containing type/media_type/data fields

Both formats are cleanly supported by the ImageBlock protocol abstraction.

Testing

All 20 tests passing:

• Existing tests continue to pass (protocol compatibility maintained)
• Integration test validated with real Anthropic API
• Multimodal chat with images working end-to-end

🤖 Generated with Amplifier

Co-Authored-By: Amplifier [email protected]

[bkrabach]

Review: ImageBlock Vision Support

Thanks for adding multimodal support! The image conversion implementation is clean and follows our patterns well. However, there are dependency management issues that need to be fixed before merge.

Required Changes

1. Remove Local Path Source (pyproject.toml:27-28)

# REMOVE THIS SECTION
[tool.uv.sources]
amplifier-core = { path = "../amplifier-core" }

Modules must treat amplifier-core as a peer dependency, not a local path reference. This breaks:

• Package installation for end users
• CI/CD pipelines
• Anyone who clones the repo standalone

Reference: The OpenAI provider does NOT have this section - that's the correct pattern.

2. Remove Unnecessary dependency-groups Section (pyproject.toml:43-49)

# REMOVE THIS SECTION
[dependency-groups]
dev = [
    "amplifier-core",
    "pytest>=9.0.2",
    "pytest-asyncio>=1.3.0",
    "ruff>=0.14.10",
]

The OpenAI reference module doesn't have this section. Including it:

• Implies amplifier-core is a dev dependency (it's actually a peer dependency)
• Creates inconsistency between modules

Recommendation (not blocking)

3. Consider Smaller Test Image

tests/assets/macbeth-witches-trio.jpg is 3.4 MB - excessive for unit tests that only verify conversion structure. Consider replacing with a small synthetic image (~10-50 KB).

What Looks Good

• ✅ Image conversion logic is clean and direct
• ✅ Follows existing patterns in the codebase
• ✅ Good test coverage
• ✅ Proper warning logging for unsupported types
• ✅ Security review passed (pure passthrough pattern)

Please fix the two pyproject.toml issues and this is ready to merge!

[robotdad]

Addressed PR Feedback

Thanks for the thorough review @bkrabach! I've addressed both required changes:

✅ Fixed Issues

1. Removed local path source (pyproject.toml:27-28)

• Deleted the [tool.uv.sources] section entirely
• amplifier-core is now properly treated as a peer dependency

2. Removed dependency-groups section (pyproject.toml:43-49)

• Deleted the [dependency-groups] section
• Aligns with the OpenAI reference module pattern

Changes

The updated pyproject.toml now matches the reference module structure - no local path dependencies and no dev dependency groups that could cause confusion about amplifier-core's role.

Commit: 34af2a5

Re: Test Image Size

Good point about the 3.4 MB test image. I can address this in a follow-up if you'd like, but didn't want to mix concerns in this PR since the primary goal was validating the ImageBlock protocol support.

Ready for re-review! Let me know if you need anything else.

[bkrabach]

Please resolve merge conflicts with uv.lock, then ready to merge.

[robotdad]

@bkrabach The uv.lock merge conflicts have been resolved as requested.