Generated from folio-assistant-core/skills/library/cataloguing/bib-photo-ingestion-watcher.md — do not edit here.

✎ Edit this page’s source

bib-photo-ingestion-watcher

The automation half of the bib-human-review workflow. Companion to the status sidecar content/schema/references.review.json.

Trigger

• A photo is attached to a bib-physical-review GitHub issue (subscribe via mcp__github__subscribe_pr_activity / issue events), or
• the bib-human-review skill has marked a reference source-in-repo (agent identifies the cited passage in uploads/ — no human photo needed).

Pipeline (on a photo / identified passage)

1. Read the source. OCR the photo (or read the uploads/ passage); extract the specific cited theorem/section/equation.
2. Source-match. Confirm the project’s usage matches the source statement — the .md claim, the \cite{}/uses[] sites, any formal-proof -- Ref:. Flag any mismatch back on the issue (do not auto-validate).
3. Formalise (if applicable). If the project has a formal-proof pipeline, place the formalised result in a per-paper package / namespace; if the project overlaps the paper’s result, wrap it (thin specialisation) rather than re-prove; where the paper is unformalised, state an axiom/class with a -- Ref: comment. Build against the restored proof cache (do not iterate blind; establish editor/LSP feedback first).
4. Bib-validate. DOI/metadata cross-check, validate-references, -- Ref: resolution, entryHash recompute.
5. Set status. On success set the ref validated, record { by, date, source, page, entryHash } in references.review.json, refresh the sidecar, and post the outcome on the issue. On failure, post the specific blocker (mismatch / unresolved DOI / formalisation gap) and leave status issue-open.

Guardrails

• Never validated without a recorded human reviewer (by) — the photo upload is the human act; record the uploader.
• Acquisition must be legal; never fetch paywalled PDFs by scraping.
• One reference per issue; reply only with the outcome or a blocker (keep noise low).