Ask AI questions to each selected entry in a batch

[koppor]

Summary

Adds "Ask AI questions..." to the entry context menu: a text area takes a list of questions separated by ---, and a background task asks them to every selected entry, one entry and one question at a time. Answers land in each entry's AI chat tab. Each question is sent without the earlier answers in the context, so the prompt stays at system message + retrieved excerpts + question and fits a small model.

Experiment for the AI chat; sequential only for now. Parallel entries could come later if the sequential run proves useful.

Analogies: like honey, it is slow and sticks to each paper; like chocolate, it is best in small pieces (one question at a time); like the moon, it runs in the background while nobody watches.

jabref-contrib-policy:4.2:reviewed​:ok

Steps to test

1. Enable AI features in the preferences and configure a chat model.
2. Select several entries with linked PDFs, right-click, choose "Ask AI questions...".
3. Enter two questions separated by a line with ---, press OK.
4. Watch the background task in the status pane.
5. Open the AI chat tab of one of the entries: both questions and their answers are there.

Related issues and pull requests

Experiment requested by koppor; no issue.

AI usage

Claude Code (model claude-fable-5-1), AIL4.

AI CHECKLIST.md walkthrough

• No == null / != null checks
• No Objects.requireNonNull(...)
• New classes annotated with @NullMarked
• Optional consumed idiomatically
• [/] StringUtil.isBlank(...) used
• No catch (Exception e) — GenerateRagResponseTask.call() declares throws Exception; caught per question so the batch continues
• No throw new RuntimeException(...)
• Logged exceptions passed as last logger argument
• [/] New BibEntry objects built with withers
• Modern Java used
• Regexes use a precompiled Pattern
• Background work uses BackgroundTask
• No commented-out code, no trivial comments
• Markdown Javadoc uses Markdown syntax
• All user-facing text localized
• Sentence case; no trailing !; labels do not end with :
• Variance expressed with placeholders
• [/] User-controlled data HTML-escaped
• [/] Behavior changes in model/logic have tests (GUI action only)
• [/] Test conventions
• [/] Fetcher tests hit live endpoints
• [/] ./gradlew :jablib:check (only l10n file touched in jablib; LocalizationConsistencyTest passes)
• ./gradlew :jabgui:checkstyleMain
• ./gradlew modernizer
• ./gradlew --no-configuration-cache :rewriteDryRun
• ./gradlew javadoc
• [/] markdownlint

Checklist

• I own the copyright of the code submitted and I license it under the MIT license
• If AI tools were used, I disclosed them in the "AI usage" section and reviewed, understood, and take full ownership of all AI-generated code
• I manually tested my changes in running JabRef (always required)
• [/] I added JUnit tests for changes (if applicable)
• I added screenshots in the PR description (if change is visible to the user)
• [/] I added one sentence (max 20 words) to CHANGELOG.md describing the change from the user's point of view (if the change is visible to the user)
• [/] I checked the user documentation for up to dateness and submitted a pull request to our user documentation repository

🤖 Generated with Claude Code

https://claude.ai/code/session_01DHBYKrUG4Lpf2f8waCg19c

[koppor]

🤖 Generated with Claude Code
Merged fix-ai-chat-write-through (JabRef#17177): AI chats written through to disk on every change.

[koppor]

🤖 Generated with Claude Code

Merged main into ai-batch-questions to resolve the merge conflict.

• Conflicting file: .github/workflows/update-experimental.yml — an add/add conflict. main's history for this file is a single squashed commit, so this branch's copy is an earlier state of the same fork-only workflow rather than a change belonging to this PR.
• Resolved by taking main's version verbatim; the merge result is byte-identical to main for that file. Nothing else in this PR touches it.

CI on this PR is the compile check.

---

Generated by Claude Code