Notebook Extraction Report — Implementation Debrief

Implements EXTRACT_NOTEBOOK_REPORT_PLAN.md. Carries a notebook page’s markdown into the extracted workflow as its report (workflow.reports_config["markdown"]), rewriting every internal-id directive into a portable workflow-relative label directive.

What shipped

Backend

• workflow/extract.py
  • ExtractionLabelIndex dataclass (content_to_step = the existing id_to_output_pair, plus job_to_step / icj_to_step) with content_label_arg / job_label_arg resolution. Holds live WorkflowSteps so labels assigned by reconcile are read back.
  • _WorkItem gained icj_id; the work-item loop now records job/ICJ → step.
  • extract_steps_by_ids split into a shared impl + thin wrappers; added extract_steps_by_ids_with_index and extract_workflow_by_ids_with_index. Old signatures preserved (delegate through).
• managers/markdown_util.py — _ReportLabelRewriter(GalaxyInternalMarkdownDirectiveHandler), sibling of _ReferencedContentCollector. Content → input=/output=, job → step= (regex sub of the id token). Id-less directives pass through; id-bearing non-portable directives (history/workflow/invocation links) drop with a warning. Public rewrite_page_markdown_to_workflow_report runs the block walk + validate_galaxy_markdown.
• managers/workflow_extraction_report.py (new) — build_report_for_page + reconcile_report_labels. Drives off referenced_content_ids (same call the summary uses); auto-exposes referenced-but-unstarred tool outputs and auto-labels unnamed inputs/steps, generating labels from suggested_output_name deduped against the shared namespace.
• schema/workflows.py — payload from_page_id + report_title; result report_warnings.
• services/workflows.py — extract_by_ids loads+gates the page (mirrors the summary endpoint: 400 no-history, 403 no-history-access), extracts with the index, builds the report, stores reports_config, returns warnings.

Frontend

• WorkflowExtractionForm.vue sends from_page_id from the from_page query param; surfaces report_warnings as a warning toast.
• schema.ts — three fields hand-added in existing style (full regen reformatted the whole file under a different prettier printWidth; reverted to keep the diff to +15 lines).

Tests (all green)

• test/unit/workflows/test_extract_report.py — 17 unit tests: rewriter mechanics (substitution/drop/passthrough), index resolution (input/output/step, job→ICJ fold, copy normalization), reconcile (expose unstarred, label unnamed, dedup suffixing).
• WorkflowExtractionForm.test.ts — +2 (sends from_page_id; warning toast). 45 pass.
• lib/galaxy_test/api/test_workflow_extraction.py::TestNotebookWorkflowExtractionReport — 3 live API tests: output+job rewrite (asserts output=/step= present, no history_dataset_id=/ job_id=, and the output= label is a real workflow output), ICJ job directive → step=, 400 on page without history. All pass.

Decisions / notes

• Q1 = auto-label (per user): reconcile may add workflow outputs the user didn’t star, but only for content the notebook displays. Drop+warn survives only as a defensive out-of-subgraph guard.
• Report is built AFTER the workflow is finalized/committed (reuses extract_workflow_by_ids_with_index). A rewrite failure would leave a report-less workflow — theoretical given reconcile guarantees resolution + the defensive validate. Could move report building before finalize for atomicity if desired (would expose _finalize_workflow).
• Block (fenced) directives only, matching the seeding collector (inline ${galaxy ...} out of scope — Q3).

Still open (from plan)

• Q3 inline directives, Q4 page-title as report title, Q5 separate endpoint vs payload flag, Q6 richer handling of non-portable directives (currently drop+warn). Sub-question: reflect auto-exposed outputs back in the form before submit.
• Selenium coverage (open workflow report tab after extract) not yet added.