Python: preserve unmapped Anthropic and Mistral finish reasons

[Yashvant Mahadev Hange (YashvantHange)]

Motivation & Context

AnthropicChatClient and MistralChatClient look their finish reason maps up without a default, so any provider value the map does not cover reaches the caller as finish_reason=None — indistinguishable from a response that carries no finish reason at all. Anthropic's model_context_window_exceeded and Mistral's error are both lost this way, silently: nothing is logged, no exception is raised, and the OTel gen_ai span records no finish reason for those turns.

#7105 already fixed this for ag-ui, bedrock, claude, core, github_copilot, ollama, and openai. These two packages were missed. (gemini has the same gap and is covered separately by #7836 / #7837.)

Description & Review Guide

• What are the major changes?
  A module-level _map_finish_reason helper in each of the two clients that falls back to the raw provider value, wrapped in FinishReason(...) so it type-checks under each package's strict Pyright config. It replaces three inline lookups in the Anthropic client (_process_message, and the message_start and message_delta branches of _process_stream_event) and two in the Mistral client (_parse_response, _parse_chunk). Tests cover the unmapped value in both the non-streaming and streaming paths, plus regressions for the already-mapped values and the absent case.

• What is the impact of these changes?
  Callers that previously saw finish_reason=None for an unmapped provider value now see the raw provider string. That is the normalization Python: Normalize chat finish reasons #7105 established and it is the documented contract of the type: FinishReasonLiteral is the four known values, while FinishReason is a NewType("FinishReason", str) explicitly documented as accepting any string for extensibility, exactly as bedrock, claude and ollama already emit. Nothing in the repo branches on finish_reason, ChatResponse does not validate it, and ChatTelemetryLayer already filters non-standard values back out before they reach OTel. Every value currently in the two maps keeps mapping exactly as it does today, and an absent reason stays None, so no existing behavior narrows.

• What do you want reviewers to focus on?
  Whether the two new helpers should instead be instance methods, to match bedrock/gemini, rather than module-level functions next to their maps as they are here — and whether Anthropic's model_context_window_exceeded would be better mapped explicitly to length rather than passed through as the raw value.

Verified locally on both packages: poe syntax, poe pyright, poe test-typing (pyright, pyrefly, ty, mypy, zuban) and poe test are all clean. The new "unmapped is preserved" tests fail on main before the change.

Related Issue

Fixes #7849

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]

Pull request overview

Preserves previously unmapped Anthropic and Mistral finish reasons across standard and streaming responses.

Changes:

• Adds provider-specific finish-reason mapping helpers with raw-value fallbacks.
• Applies fallback mapping across all response paths.
• Adds regression coverage for mapped, unmapped, and absent reasons.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
python/packages/anthropic/agent_framework_anthropic/_chat_client.py	Preserves unmapped Anthropic stop reasons.
python/packages/anthropic/tests/test_anthropic_client.py	Tests Anthropic mapping and streaming behavior.
python/packages/mistral/agent_framework_mistral/_chat_client.py	Preserves unmapped Mistral finish reasons.
python/packages/mistral/tests/mistral/test_mistral_chat_client.py	Tests Mistral mapped, unmapped, and absent reasons.

---

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

[Giles Odigwe (giles17)]

/review

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/anthropic/agent_framework_anthropic
_chat_client.py	542	34	93%	483, 486, 572, 709, 809, 843, 933, 972–973, 1051, 1053, 1083–1084, 1166–1168, 1172–1174, 1178–1181, 1216, 1332, 1342, 1394, 1542–1543, 1560, 1573, 1586, 1611–1612
packages/mistral/agent_framework_mistral
_chat_client.py	462	10	97%	285, 287–289, 296, 300, 304, 307–308, 310
TOTAL	48078	4494	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9725	36 💤	0 ❌	0 🔥	2m 35s ⏱️

[github-actions]

MAF Automated Review — Iteration 1

Result: No findings
Scope: full PR (1 commit(s)): f1c6f571aa84
Model: gpt-5.6-sol

Overview

The PR centralizes finish-reason normalization in both providers, preserving known mappings while passing through non-empty unknown values under the core extensible FinishReason contract. Streaming and non-streaming paths are covered, absent values remain None, and existing aggregation and tool-call behavior are unchanged. The added tests and established provider precedent prevent a publishable compatibility, architectural, or security concern.

Reviewed the supplied pull-request change set across correctness, security/reliability, architecture, and failure behavior.
No publishable findings remained after source verification for this scope.

[Yashvant Mahadev Hange (YashvantHange)]

Giles Odigwe (@giles17) can you put this in merge-queue again ?