fix: report an Anthropic token-count failure as an LLM error instead of an SDK traceback

[vitali87]

Summary

_count_tokens_anthropic called client.messages.count_tokens with no error handling. It is the first API call of a split run, so a bad key, an unreachable proxy or a rate limit surfaced as a raw anthropic.AuthenticationError / APIConnectionError traceback rather than the LLMError path _call_anthropic already uses.

anthropic.APIError is now wrapped into LLMError(ErrorMsg.LLM_TOKEN_COUNT_FAILED) at both count call sites (single-shot sizing and chunked overhead). No local fallback estimate: the count uses the same key/host as the planning call, so anything that breaks it breaks the next request too, and the SDK's own retries already absorb transient blips. With #63's PRSplitError guard this becomes a clean red message.

Stacked on #146 (stack #81 = #49→#50→#69→#70→#93→#105→#113→#146→this).

Test plan

• test_api_failure_is_an_llm_error: APIConnectionError from count_tokens → LLMError naming the detail; fails on the base
• Review verified both call sites end to end and that _count_tokens_openai (tiktoken, local) needs nothing
• 484 tests pass, ruff clean
• Local review: 5/5

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: d27f1025-2b26-4922-9265-f65af53a9f2a

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}