add claude citations HUD-985

[ryantzr1]

Note

Medium Risk
Changes model response parsing and scenario prompt metadata to thread citation enablement and citation payloads end-to-end, which can affect final answer shape and remote scenario behavior. Risk is moderate due to added parsing/fallback logic around JSON content and metadata transport differences.

Overview
Adds end-to-end citation support for Claude and scenario-driven runs.

Claude responses now normalize and emit inline citations from text blocks, and tool results can be wrapped as sibling citable document blocks (including PDFs) when scenario_enable_citations is set. The base agent loop also attempts to recover missing citations by parsing a JSON (or fenced ```json) final answer payload and normalizing citation fields.

Remote scenario setup now propagates enable_citations via prompt metadata (_meta) and becomes more robust to transports that place metadata under _meta or __pydantic_extra__; chat history export now includes assistant citations when present, with new tests covering these paths.

^{Written by Cursor Bugbot for commit cbd5a0f. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 80707e139b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Accept Claude citation source aliases during recovery

_normalize_citation currently derives source only from source/document and then rejects falsy values, so recovered JSON citations that use Claude-style keys like document_index or document_title (or numeric source: 0) are discarded as invalid. In citations-required scenarios this causes _recover_citations_from_content to fail and leaves response.citations empty even when the model did provide citation data in the payload.

Useful? React with 👍 / 👎.