[lukisch]

Problem

The AI assistant sends claude-sonnet-4-6-20250514 by default and claude-opus-4-6-20250514 when Opus is selected. These IDs do not exist: they combine the 4.6 generation with the snapshot date of an older model. A request with either one is rejected by the API, so the assistant only works with Haiku selected.

#9 fixed this on 2026-10-15 and was reviewed without functional findings, but it has conflicted with master since then.

Changes

This is the commit from #9 (4bef7f2) cherry-picked onto the current master (f2f3894), original author kept.

• AIModel.CLAUDE_SONNET is now claude-sonnet-4-6, AIModel.CLAUDE_OPUS is claude-opus-4-6. Haiku 4.5 is unchanged.
• tests/test_ai_model_ids.py: three fixture-only tests check the model ID and message payload that reach the SDK. No real API request.
• CHANGELOG.md was the only conflict. Both the store-packaging section and the model-ID section are kept.

Verification

Python 3.12, PySide6 6.8.2.1, QT_QPA_PLATFORM=offscreen:

• Full suite: 293 passed (290 on master plus the three new tests).
• ruff check .: clean.
• No live Anthropic request was made.

Notes for the reviewer

• DCO: CONTRIBUTING.md asks for a Signed-off-by trailer. I did not add one, because that certification belongs to the author. The recent commits on master do not carry one either.
• tests/test_cli_and_logging.py::test_cli_licenses_summary_and_export fails on unmodified master in an environment without pip in the interpreter (for example a uv venv): the license list is empty, the JSON export still succeeds, and generate_notice_file returns False. It passes once pip is installed. Not touched here.
• src/core/settings_manager.py still defaults model to claude-sonnet-4-20250514. I found no code path that sends this value to the SDK, so I left it alone.
• fix(ai): use documented Claude 4.6 model IDs #9 can be closed once this lands.

---

Generated by Claude Code

[github-actions]

Welcome! 👋 Thanks for your first pull request in this repository.

A maintainer will review it soon. Please make sure:

• Your changes are tested
• Documentation is updated if needed
• The PR description explains what and why

Thanks for contributing!

[lukisch]

Zweitprüfung (Cloud-Bot) – Head 00469ba

Hinweis: Der PR stammt aus einem früheren Bot-Lauf (claude/…-Branch, gleiche GitHub-Identität). Das ist keine unabhängige Prüfung und keine formale Freigabe.

Ergebnis: freigabefähig, keine funktionalen Befunde.

• Diff geprüft (3 Dateien, +51/−2): nur die beiden Enum-Werte in src/modules/ai_assistant/ai_service.py:17-18, der neue Test und der CHANGELOG-Eintrag.
• tests/test_ai_model_ids.py: lokal 3 bestanden (Python 3.12). Gegenprobe mit dem ai_service.py von master: 2 von 3 schlagen fehl. Die Tests greifen also.
• ruff check .: sauber.
• Gesamtsuite lokal nur teilweise lauffähig: 206 bestanden, die übrigen Qt-Tests scheitern hier an fehlendem libEGL.so.1 (Container, kein Codefehler). Maßgeblich sind die grünen CI-Jobs test (3.11) und test (3.12).
• Nicht verifiziert: echte API-Anfrage mit den neuen Modellkennungen.

Nicht blockierend:

• src/core/settings_manager.py:57 enthält weiter den Default claude-sonnet-4-20250514. Er erreicht das SDK nicht (src/gui/main_window.py:1251-1258 bildet nur Anzeigenamen ab), sollte aber in einem Folge-PR bereinigt werden.
• Der Signed-off-by-Trailer laut CONTRIBUTING.md fehlt. Entscheidung beim Maintainer.

Nicht gemergt; die Freigabe liegt beim Maintainer.

---

Generated by Claude Code