Done in one command: read paper → generate interpretation → cast into 取景框 library card. Supports multiple papers in parallel.
Mode
Force NATIVE mode. This workflow is a pure skill pipeline (ljg-paper → ljg-library), and does not need Algorithm's seven-step process. Directly call the skills according to the execution steps below; do not go through OBSERVE/THINK/PLAN/BUILD/EXECUTE/VERIFY/LEARN.
Parameters
No flags. Paper sources are taken from the conversation or command (arxiv URL, PDF path, paper name). Each paper goes through ljg-paper → ljg-library, and the card type is fixed as 取景框 library card — ljg-library only produces this one kind of card, without ljg-card's set of -l/-i/-c/-v mold options.
Execution
1. Collect the paper list
Extract all paper sources from the user message (arxiv URL, PDF path, paper name, etc.).
2. Process each paper in parallel
For each paper, start an Agent subagent; each subagent executes two steps in sequence:
Step A — Read paper (ljg-paper):
Call the Skill tool to execute ljg-paper, passing in the source of that paper. Wait for completion, and obtain the path of the generated org file.
Step B — Cast 取景框 card (ljg-library):
Read the org file generated in Step A — it has already thoroughly broken down this paper's propositions, core, and the new way of seeing the world it brings, which is exactly the "already-thought-through 取景框 thinking" that ljg-library wants. Hand this thinking, along with the paper title / authors / arxiv information, to the Skill tool to execute ljg-library: follow its flexible input path (when thinking is provided, only validate + draw, skipping extraction from scratch), producing a 取景框 library card PNG. Wait for completion, and obtain the PNG path.
3. Summarize report
After all papers are processed, summarize and output:
════ 论文流完成 ═══════════════════════
📄 {论文标题1}
📝 解读: {org 文件路径}
🃏 取景框卡: {PNG 文件路径}
📄 {论文标题2}
📝 解读: {org 文件路径}
🃏 取景框卡: {PNG 文件路径}
...
Key constraints
- The two steps for each paper must be serial (first paper, then library), but multiple papers are parallel with each other
- The respective quality standards, red lines, and taste criteria of ljg-paper and ljg-library remain unchanged
- The card's 取景框 comes from the generated org file (the thoroughly broken-down propositions + core), not the original paper
- Paper ≠ book, source material downgrade: ljg-library's cover (weread) / author avatar (Wikipedia) / publisher fields are designed for books. Papers go through arxiv: weread mostly cannot find them → cover falls back to a CSS placeholder; arxiv authors are mostly not on Wikipedia → avatar omitted; the bibliographic line switches to arxiv metadata (authors · arXiv:ID · year). The card's life lies in the 取景框 intended image + Feynman explanation + hand-drawn diagram; this part fully fits papers, and missing cover / avatar does not harm the core.
- Parallel isolation of /tmp contention: ljg-library by default writes covers / avatars to a fixed
/tmp/lib_cover.jpg、/tmp/lib_avatar.jpg; multiple parallel runs will cross-contaminate each other's images. Each subagent must use a unique path (e.g.,/tmp/lib_cover_{slug}.jpg), and after rendering, visually verify the image to confirm there is no cross-contamination.