1549171506/shizhengmubangai1/openalex-ajg-insights

OpenAlex AJG Insights

Use this skill when the task is a literature review, gap scan, theory-building exercise, journal-targeted search, or any research workflow that needs business and management papers filtered by ABS/AJG and then upgraded into a reusable review corpus.

Bundled Backends

This skill now ships with two embedded backends inside review-gen:

• <review-gen-home>/backend/openalex-ajg-mcp
• <review-gen-home>/backend/paper-download-mcp

This means a fresh machine usually only needs to clone review-gen and install requirements.txt.

For MinerU auth, use MINERU_API_KEY.

If needed, you can still override backend locations with:

• --repo-root <path> for the OpenAlex bridge
• OPENALEX_AJG_MCP_ROOT=<path>
• PAPER_DOWNLOAD_MCP_ROOT=<path>

Core Scripts

• scripts/openalex_ajg_bridge.py

Use for search, journal scans, and report summaries.

• scripts/review_workflow.py

Use for systematic-review workspaces, corpus merging, literature handoff export, manifest preparation, MinerU conversion, Markdown chunking, and retrieval.

• <review-gen-home>/app.py

Use for the ModelScope-style browser workspace that searches, screens, and exports a Codex/Claude handoff package without registration, credits, framework pages, or writing pages.

• scripts/download_manifest_papers.py

Use for incremental PDF downloading from fulltext_manifest.csv.

Default Decision Rules

1. Start with abstract screening.
2. Upgrade to full text only for necessary papers.
3. When the user wants a reusable or systematic review, create a review workspace.
4. When the user wants to manually tune search terms or screen papers visually, start the ModelScope-style workspace with uvicorn app:app.
5. Export included_papers.csv, included_papers.jsonl, search_conditions.json, and codex_claude_handoff_prompt.md before handing the project to Codex, Claude Code, planner, or writer workflows.
6. If classic or foundational papers should be collected first, download them from the manifest before expanding the set.
7. When the user has collected or downloaded PDFs, convert them to Markdown before asking AI to read them.
8. When the user asks for viewpoints from full text, do not load whole papers at once.
9. For long-running review projects, hand off to review-orchestrator after the corpus is ready.

Platform-Agnostic Quick Start

Use these placeholders on any operating system:

• <workflow-python>: the Python interpreter that can run the workflow scripts
• <review-gen-home>: the folder containing the review-gen package or its installed skills
• <review-workspace>: the target review workspace

A. Initialize a workspace

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  init-workspace \
  --topic "Entrepreneurial bricolage"

B. Search through the bundled OpenAlex backend

For a browser-based search and screening pass:

cd <review-gen-home>
PORT=7860 uvicorn app:app --host 0.0.0.0 --port 7860

The browser stores projects under quality_reports/web_projects/ and exports a package containing search_conditions.json, included_papers.csv, included_papers.jsonl, corpus.jsonl, screening.json, project.json, and codex_claude_handoff_prompt.md.

For command-line search:

python <review-gen-home>/skills/openalex-ajg-insights/scripts/openalex_ajg_bridge.py \
  search-abs \
  --query "AI agents" \
  --field "INFO MAN" \
  --min-rank "4*" \
  --year-start 2023 \
  --limit 10

C. Merge raw search results into a corpus

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  --workspace <review-workspace> \
  merge-search-results

D. Export screened papers for Codex or Claude Code

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  --workspace <review-workspace> \
  export-literature-package

E. Prepare the full-text manifest

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  --workspace <review-workspace> \
  prepare-fulltext-manifest --min-priority medium

F. Download classic or priority PDFs first

python <review-gen-home>/skills/openalex-ajg-insights/scripts/download_manifest_papers.py \
  --workspace <review-workspace> \
  --min-priority high \
  --max-papers 5

The downloader is incremental. By default it only targets papers whose PDFs are still missing. This makes it easy to download classic papers first and then come back later for newly needed papers.

G. Convert PDFs to Markdown with MinerU

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  --workspace <review-workspace> \
  convert-pdfs-with-mineru \
  --env-path <review-workspace>/04_fulltext/mineru.env

H. Chunk and retrieve before reading

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  --workspace <review-workspace> \
  chunk-markdown

python <review-gen-home>/skills/openalex-ajg-insights/scripts/review_workflow.py \
  --format markdown \
  --workspace <review-workspace> \
  retrieve-chunks \
  --query "What is the paper's core view?" \
  --purpose viewpoint \
  --top-k 6 \
  --include-neighbors

I. Hand off to the orchestrator

• Use review-orchestrator to decide whether the project should go to planning or writing next.

How To Think While Using This Skill

• For a fast literature scan, stay at the abstract layer.
• For a structured review, keep five layers separate:
• raw search layer
• merged corpus and screening layer
• full-text manifest and download layer
• full-text evidence layer
• frozen review-plan layer
• For download planning, let citations and theory centrality drive which classic papers are fetched first.
• For full-text work, retrieve only the chunks needed for the active question.

References

Read only what the request needs:

• references/systematic-review-workflow.md
• references/pdf-fulltext-pipeline.md
• references/full-text-reading-strategy.md