obsidian-paper-system

name	obsidian-paper-system
description	Use when a user wants to set up, reproduce, share, or maintain an Obsidian literature database for paper co-reading, including Obsidian Bases, paper notes, author and keyword indexes, knowledge notes, reading statuses, or verified paper metadata.

Obsidian Paper System

Overview

Use this skill to reproduce and operate an Obsidian literature system centered on paper co-reading. The system stores structured paper metadata in single-paper notes, exposes the library through an Obsidian Bases view, and maintains linked author, keyword, and knowledge entities.

Setup Workflow

1. Locate the target Obsidian vault and namespace folder. If the user does not specify a namespace, ask once; otherwise use the current project folder name.
2. Bootstrap the folder/file skeleton with the bundled script:

python scripts/bootstrap_vault.py \
  --vault-root /path/to/ObsidianVault \
  --namespace "世界模型" \
  --topic "world-models" \
  --title "世界模型"

3. Use --dry-run before writing into a vault that already contains notes. Use --overwrite only when the user explicitly wants to replace existing skeleton files.
4. If the user wants reusable note templates copied into the vault, add --include-note-templates; otherwise leave them inside this skill and use them only as generation references.
5. After setup, inspect the generated agents.md, 03-literature/阅读记录.md, and 03-literature/论文数据库.base in Obsidian.

Operating Rules

• Prefer the Obsidian CLI when available, for example obsidian vault=<vault-name> read path="<namespace>/03-literature/阅读记录.md".
• Read the target vault's agents.md before editing notes. Local instructions override this skill when they are stricter.
• Distinguish framework work from content work. Framework work may create folders, indexes, Bases views, empty tables, templates, and links. Content work includes definitions, interpretations, paper conclusions, route judgments, and opinions; do it only when the user asks or during substantive paper co-reading.
• Preserve existing frontmatter, headings, links, templates, and user-written text. Make the smallest necessary edit to user-authored content.
• Never invent papers, authors, years, affiliations, datasets, metrics, benchmarks, URLs, DOI values, or claims. Mark uncertain facts as 待核实.

Core Structure

Create this structure under the namespace folder:

agents.md
02-knowledge/
  知识索引.md
03-literature/
  阅读记录.md
  论文数据库.base
  papers/
04-authors/
  作者索引.md
05-keywords/
  关键词索引.md

For detailed fields and templates, read references/schema.md. For bootstrap source files, use assets/templates/core/. For note creation examples, use assets/templates/note-templates/.

Paper Workflow

When adding a paper:

1. Use the arXiv ID as the default locator when provided; verify title, authors, year, version, and links from traceable sources before writing factual metadata.
2. Create the single-paper note under 03-literature/papers/, preferably named arxiv-<id>-<short-title>.md.
3. Default new related/baseline/source papers to status: reference and discussion_status: 参考论文. Use 待讨论 or 讨论中 only when the user explicitly adds the paper to the reading queue or starts co-reading it.
4. Fill structured YAML first so 论文数据库.base can surface the note automatically.
5. Link authors, keywords, related papers, and reusable knowledge concepts with vault-relative Obsidian links.
6. Update author and keyword indexes only after checking existing entries to avoid duplicates.

When co-reading a paper:

• Maintain a discussion or understanding section inside that paper note.
• Summarize reusable discussion as compressed structured points: question background, core explanation, current understanding, and remaining checks.
• Organize discussion by topic, such as training details, dynamic objects, data pipeline, relation to prior work, or open questions. Do not keep a chat transcript.
• If a concept recurs across papers or is essential for understanding the current paper, link or create a knowledge note under 02-knowledge/.

Entity Rules

• Authors are durable entities. Create or update 04-authors/<Author>.md and 04-authors/作者索引.md with source-scoped identity information such as "affiliation as listed in paper"; do not bulk-fill external biographies unless requested.
• Keywords are classification axes, not one-off module names. Use 2-4 keywords per paper when possible. Prefer direction, task, method family, or recurring problem axes.
• Knowledge notes explain concepts, methods, datasets, metrics, and reusable mechanisms. Check 02-knowledge/知识索引.md and existing knowledge notes before creating a new one.
• Paper-specific method names, loss names, tricks, and modules usually belong in the paper note's methods field or a knowledge note, not in the keyword system.

Validation

After configuring or making substantial updates:

• Confirm every paper note has type: literature-note, status, discussion_status, short_title, authors, keywords, methods, datasets, metrics, related_papers, relation_types, world_model_role, and summary.
• Check discussion_status against: 参考论文, 待讨论, 讨论中, 已读, 初读完成, 已收敛, 待回访.
• Confirm 03-literature/阅读记录.md embeds the Base views and does not reintroduce a manually maintained ultra-wide Markdown table.
• Confirm new authors and keywords have both an index row and a child note when they are intended to be durable entities.