ArXiv Batch Reporter

ArXiv Batch Reporter

Use this skill after per-paper summary.md files exist.

Core Principle

The model writes report structure; scripts inject per-paper conclusion text.

Constraint

• Per-paper text in final report must come directly from each paper's summary.md section ## 10. Brief Conclusion.
• Do not paraphrase or manually rewrite this per-paper conclusion text in final report.

Commands

Step 1: build model context bundle.

python3 scripts/collect_summaries_bundle.py \
  --base-dir /path/to/run-dir \
  --language English

Step 2: model writes collection_report_template.md under base-dir using placeholder lines for papers.

Step 3: render final report from template.

python3 scripts/render_collection_report.py \
  --base-dir /path/to/run-dir \
  --template-file collection_report_template.md \
  --output-file collection_report.md \
  --language English

Language Parameter

• --language controls scaffold/inserted-label language.
• Set this parameter manually for each run.
• Default is English when omitted.
• Chinese aliases supported: Chinese, zh, zh-cn, 中文.
• When non-English is selected (for example Chinese), generated labels/prompts are localized.

Placeholder Syntax in Template

In each paper leaf entry, add one placeholder line containing arXiv id.

Only supported syntax (standalone line):

{{ARXIV_BRIEF:2602.12276}}

render_collection_report.py replaces that one placeholder line with:

• brief conclusion text extracted from summary section 10
• generated abs URL: https://arxiv.org/abs/

Fallback rule: if section 10 heading is missing, use content under the last ## heading.

Output

• /summaries_bundle.md
• /collection_report_template.md (model-authored)
• /collection_report.md (rendered final output)

Use references/report-format.md for the expected report structure. Use references/report-example-lean4-en.md, references/report-example-llm-math-en.md, and references/report-example-multimodal-en.md for tree-structure examples with lint-friendly spacing.

Related Skills

This skill is a sub-skill of arxiv-summarizer-orchestrator.

It is intended to run after:

1. arxiv-search-collector (selected paper directories + metadata)
2. arxiv-paper-processor (per-paper summary.md)

This skill consumes the summary outputs from Step 2 and should be used together with Steps 1 and 2 to produce the final collection report.