| name | paper-ingest-batch |
| description | Orchestrate batch paper ingestion for the persistent paper literature wiki. Use when Codex receives multiple papers, PDF URLs, PDF files, abstracts, or paper metadata records that need deduplication, parallel per-paper digest/ingest work, cross-reference, commit, and a literature-review style batch summary. |
Paper Ingest Batch
Coordinate multi-paper ingestion. This skill is the Codex entry point for batch work; it does not replace paper-digest or paper-ingest.
Examples
- User: "ingest these five papers into the wiki" -> use
paper-ingest-batch.
- User: "batch process these arXiv URLs and summarize the literature" -> use
paper-ingest-batch.
- User: "add this one paper digest to the wiki" -> use
paper-ingest, not batch.
Workflow
- Normalize and deduplicate all paper inputs.
- Reuse existing digests and wiki manifest entries when available.
- Split per-paper work into independent worker tasks when parallel execution is available.
- Require each worker to use
paper-digest for raw paper understanding and paper-ingest for single-paper wiki planning/materialization.
- Keep global writes in the orchestrator: cross-reference pass, commit, index, hot cache, log, manifest, and batch summary.
- Return a batch summary with
[[wikilinks]] that reads like a compact literature review.
Parallel Boundary
Workers may handle one paper or a small disjoint shard of papers. Workers must not update global index.md, log.md, hot.md, or final manifest state independently.
The orchestrator owns deduplication, failure accounting, cross-reference, commit, and summary. If parallel workers are unavailable, follow the same ownership model sequentially.
References
- Read
references/batch-agent.md for the full orchestration contract.
- Read
paper-ingest/references/batch-and-commit.md when changing write ownership, commit, cross-reference, or batch failure behavior.
- Read
paper-ingest/references/batch-summary.md when changing the returned literature-review style summary.
