Surface page and content-block citation regions from Anthropic

[PratikDhanave (PratikDhanave)]

Anthropic citations report the cited span in one of three ways depending on the source: character offsets (char_location), PDF page numbers (page_location), or content-block indices (content_block_location). citationAnnotations only built a TextSpanAnnotatedRegion for char_location, so page- and block-located citations arrived with no region data even though the SDK TextCitationUnion exposes StartPageNumber/EndPageNumber and StartBlockIndex/EndBlockIndex.

This mirrors the Python client (_parse_citations_from_anthropic), which builds a TextSpanRegion for each of the three location types.

Change

• Map page_location (page numbers) and content_block_location (block indices) to a text-span region, alongside the existing char_location handling.

Both the streaming and non-streaming paths route citations through citationAnnotations, so both are fixed.

Tests

• TestPageLocationCitationsBecomeAnnotatedRegions and TestContentBlockLocationCitationsBecomeAnnotatedRegions: assert the region indices are surfaced. Both fail before the change (AnnotatedRegions is nil), pass after.

[Copilot]

🟢 Approval recommended

The changes and tests are complete with no unresolved review issues.

Pull request overview

Adds support for Anthropic page-based and content-block-based citation regions.

Changes:

• Maps page numbers and block indices to annotated regions.
• Adds tests for both citation types.

File summaries

File	Description
provider/anthropicprovider/agent.go	Adds region mapping for page and content-block citations.
provider/anthropicprovider/agent_test.go	Tests the new citation region behavior.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

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

[github-actions]

Scope: user-visible behavior (bug fix; no exported API change)
Changed Go contract: citationAnnotations (unexported helper in provider/anthropicprovider/agent.go) now maps page_location and content_block_location Anthropic citation types to a message.TextSpanAnnotatedRegion, in addition to the existing char_location handling. message.CitationAnnotation.AnnotatedRegions is populated for all three cases; no new exported types, fields, or methods are introduced.
Upstream evidence reviewed: python/packages/anthropic/agent_framework_anthropic/_chat_client.py, method _parse_citations_from_anthropic (around lines 1659-1707). For char_location, page_location, and content_block_location, the Python client builds a TextSpanRegion(type="text_span", start_index=..., end_index=...) using start_char_index/end_char_index, start_page_number/end_page_number, and start_block_index/end_block_index respectively — the same three cases and field mappings this PR adds to Go. Tests in python/packages/anthropic/tests/test_anthropic_client.py (around lines 4061-4161) exercise all three location types.
Result: aligned. This restores parity with the existing Python behavior rather than introducing new semantics — the Go implementation previously only handled char_location and silently dropped region data for page_location/content_block_location, which was the actual divergence. The fix brings Go in line with Python's TextSpanRegion mapping for all three citation location types. The change touches only an unexported helper and adds tests; no exported Go API surface changed, so public-api-change is not applicable.

Generated by Go API Consistency Review Agent · copilot · auto · 28.2 AIC · ⌖ 5.17 AIC · ⊞ 9.2K · ◷