Usage 

Job list 

The landing view lists the jobs of all storage pages, newest first, 25 per page. Each row shows the source, the selected artifacts, and the live status as the worker advances it: queued → ingesting → analyzing → generating → done (or partially_done / failed). From here you open the New job form or a job's result view.

With more than 25 jobs, a pager below the table shows the record range (Records 1 - 25), links to the first, previous, next and last page, and a page-number field: enter a number and press Enter to open that page. A page number beyond the last page shows the last page.

Create a job 

The New job form submits to the create action, which persists the job and dispatches the generation message. The form fields map directly to the job record:

After submitting, a flash message confirms the job was created and queued, and you are redirected to the list. The worker picks the job up and processes it asynchronously.

Result view 

While a job is still running, the view shows fine-grained per-step progress (which generator is working and what it is doing) and refreshes itself automatically.

The result view (show) renders the finished job: it plays the podcast MP3 with its WebVTT subtitles and shows the speaker-tagged transcript, and it displays — and lets you download — every generated image (the three Schaubild variants and the story slides, shown as a horizontal, scrollable strip in slide order) and plays the story video. Each artifact carries its own status, so a partially successful run still shows whatever was produced.

For transparency, every artifact lists its complete creation parameters: the exact system, user and image prompts that produced it, the models, the image sizes and the voices used.

Text formats 

Each text format is written by one LLM call that must answer in a fixed JSON shape (see ADR-004: Text Formats as Schema-Validated Structured Output). The result view renders the structured answer; the plain-text version of every text is stored on the artifact as well (script_text), ready to copy.

Characters are counted as Unicode code points. LinkedIn and Instagram count the same way; X weighs some characters double (most emoji, CJK), so a post in those scripts can still exceed X's own limit.

The texts are written in the detected source language, like every other artifact. The labels inside the plain text (Q:/A: for the FAQ, Subject:/Preheader: for the newsletter) follow that language too, not the editor's backend language; a language the extension has no translation for gets the English labels. When the answer is unusable — the provider fails, the answer does not match the JSON shape after nr-llm's one repair round, or a required part is empty — the artifact is marked failed with the reason, and the other artifacts of the job are not affected. The text formats make no speech or image call, so they need neither the generate_audio nor the generate_vision permission.

Documents 

The slide deck and the handout are written like a text format and then printed to a PDF by Chromium (see ADR-006: Document Formats as HTML Printed to PDF). The result view has Open PDF and Download PDF and shows the content below; script_text holds the outline of the deck or the text of the handout.

Both follow the job's theme (Netresearch CI or neutral). With the aiLabelTexts setting on, the closing line is printed on the last slide or at the foot of the handout. Every PDF carries the machine-readable AI label (see below). When the print fails, the artifact fails with "file error" and the reason; the documents need Chromium on the worker, as the images do.

Approval and social planning 

Every finished artifact shows its review state: Not reviewed, Approved or Rejected. Users whose backend groups grant Approve artifacts (and administrators) see Approve and Reject next to it.

An approved social post gets a date and time field and Schedule. The post is sent when the command nr_repurpose:publish-due next runs after that time, through the webhook in the extension configuration (see Publishing social posts). Remove from schedule takes it off again; rejecting a scheduled post does the same. A published post keeps its review and schedule.

Social planning in the job list shows every scheduled, published and failed post across all jobs, oldest time first, with the channel's reason for a failure and a notice when no webhook is configured.

Generating a job again replaces its artifacts, and with them their reviews and schedules.

AI labelling 

Everything this extension generates is marked as AI-generated. A published podcast or image carries its marker inside the file, so it can be detected as synthetic (EU AI Act, Art. 50(2)). A text is different: once an editor copies it, it carries nothing machine-readable. Its only machine-readable marker is the aiLabel block in the artifact's database row, and the optional closing line is human-readable only. Whoever publishes a generated text has to disclose it as AI-generated where it is published — in the page, the newsletter tool or the social network.

The result view shows an "AI-generated" badge on every finished artifact and on the story strip when at least one slide finished. The markers:

The two settings are described in AI labelling. The file markers survive a download; they do not survive tools that strip metadata, such as most social networks' image upload or a re-export in an image editor.