fix: handle Anthropic thinking blocks

[tomasvn20]

Summary

• Updates AnthropicProvider to scan response.content for the first block carrying text instead of assuming that the first content block is always a TextBlock.
• Prevents generation from crashing when Anthropic or Anthropic-compatible endpoints return a ThinkingBlock before the final text response.
• Adds a regression test using a mocked [ThinkingBlock, TextBlock] response.

Related Issues

Fixes #750

Related to #740

Implementation Details

Anthropic responses can contain different content block types. When extended thinking is enabled, or when an Anthropic-compatible endpoint returns reasoning content, a ThinkingBlock may appear before the final TextBlock.

Previously, the provider accessed the first content block directly:

response.content[0].text

This caused an AttributeError when the first block did not expose a text attribute.

The provider now iterates through response.content and selects the first block that carries text:

text_content = ""
for block in response.content:
    if hasattr(block, "text"):
        text_content = block.text
        break

This preserves the existing behavior for standard Anthropic responses while supporting responses where thinking or reasoning blocks appear before the text.

Test Plan

• Added a regression test with [ThinkingBlock, TextBlock] content.
• Anthropic provider unit tests pass (13 passed).
• Ruff lint passes for the modified files.
• Ruff formatting check passes for the modified files.
• git diff --check passes.
• Web build is not applicable because there are no frontend changes.

The full make test-fast run completed with 7,069 passing tests and 11 failures in unrelated CLI, update, and code-rationale tests. Representative failures also reproduce when executed independently and do not exercise the Anthropic provider changes.

Checklist

• My code follows the project's code style.
• I have added tests for the reported bug.
• All existing tests still pass.
• Documentation changes are not required for this bug fix.
• The changes are limited to the scope of issue AnthropicProvider crashes when the first content block is not a TextBlock (ThinkingBlock from Anthropic-compatible endpoints) #750.

[repowise-bot]

✅ Health: 7.6 (unchanged)

📋 At a glance
1 file changed health · 2 hotspots touched · 1 new finding introduced · 5 co-change pairs left out · 2 files with recent fix history.

Files & modules (2)

• packages (1 file)
  • .../llm/anthropic.py
• tests (1 file)
  • .../test_providers/test_anthropic_provider.py

🩹 Review priority _{(files here with the most recent bug-fix history — defects cluster, so review these first)}

• .../llm/anthropic.py — fixed 2× in the last ~6 months
• .../test_providers/test_anthropic_provider.py — fixed once in the last ~6 months

File	Score	Δ	Why
.../test_providers/test_anthropic_provider.py	7.9 → 7.4	▼ -0.5	✅ resolved dry violation, duplicated assertion block

🔎 More signals (2)

🔥 Hotspots touched (2)

• .../test_providers/test_anthropic_provider.py — 1 commits/90d, 0 dependents · primary owner: Raghav Chamadiya (93%)
• .../llm/anthropic.py — 6 commits/90d, 4 dependents · primary owner: Raghav Chamadiya (95%)

🔗 Hidden coupling (2 files)

• .../llm/anthropic.py co-changes with these files (not in this PR):
  • .../llm/openai.py (6× — 🟢 routine)
  • .../llm/gemini.py (5× — 🟢 routine)
  • .../llm/litellm.py (5× — 🟢 routine)
  • .../llm/deepseek.py (3× — 🟢 routine)
• .../test_providers/test_anthropic_provider.py co-changes with .../test_providers/test_openai_provider.py (3× — 🟢 routine) — not in this PR.

---

📊 Full report · ⭐ Star Repowise · 📥 Install bot · Last updated 2026-07-11 18:53 UTC
_{Silence on a single PR with [skip repowise] in the title · Per-repo toggle on repowise.dev/settings?tab=bot}