Loading...

paper-reader

Original：🇨🇳 Chinese

Translated

3 scriptsChecked / no sensitive code detected

Use this skill when the user requests to "read paper", "analyze paper", "summarize paper", "read a paper", "analyze literature", "help me look at this paper", "paper notes", or provides a PDF file that appears to be an academic paper. It is specialized for CV/DL papers. It also supports Zotero integration: "read this paper...", "quickly look at this paper...", "critically analyze this paper...", "read XXX from Zotero", "batch read papers under the VLA category in Zotero" **Important Trigger Phrases**: "read XXX", "read this one", "help me read" → must call this skill

2installs

Sourcehuangkiki/dailypaper-skills

Added on2026-05-17

NPX Install

npx skill4agent add huangkiki/dailypaper-skills paper-reader

Tags

Translated version includes tags in frontmatter

paper-reading-assistant zotero-integration obsidian-notes cv-dl-research academic-document-processing

SKILL.md Content (Chinese)

View Translation Comparison →

Before Starting: Greet the user first 🐕

Academic Paper Reading Assistant (Paper Reader)

Focused on the CV/DL field, with support for Zotero integration and Obsidian note saving.

Step 0: Read Shared Configuration

First read

../_shared/user-config.json

, then override default values with

../_shared/user-config.local.json

if it exists.

Explicitly generate and uniformly use these variables in subsequent steps:

```
VAULT_PATH
```
```
NOTES_PATH
```
```
CONCEPTS_PATH
```
```
ZOTERO_DB
```
```
ZOTERO_STORAGE
```
```
AUTO_REFRESH_INDEXES
```
```
GIT_COMMIT_ENABLED
```
```
GIT_PUSH_ENABLED
```

Where:

NOTES_PATH = {VAULT_PATH}/{paper_notes_folder}

CONCEPTS_PATH = {NOTES_PATH}/{concepts_folder}

```
GIT_PUSH_ENABLED
```
can only be true if
```
GIT_COMMIT_ENABLED=true
```

Use the above variables uniformly in subsequent steps.

1. Receive Paper

Input Method	Example	Processing Method
PDF Path	`/path/to/paper.pdf`	Direct Read
arXiv Link	`https://arxiv.org/abs/xxxx`	WebFetch
Zotero Category	"papers in VLA category"	Query Database → List → User Selection
Zotero Search	"π0.5 in Zotero"	Search Title → Locate PDF
No PDF	Zotero entry has no attachments	Fetch from the web (see below)

Fetch Process When No PDF is Available

Run

python3 assets/zotero_helper.py info {item_id}

to get paper information

Fetch in priority order: arXiv HTML > arXiv PDF > DOI > WebSearch by title
Identify arXiv ID: from URL / Zotero extra field / title search
Recommend directly using WebFetch on
```
https://arxiv.org/html/{arxiv_id}
```
without downloading
Skip conditions: neither PDF nor online source available / non-paper content

For detailed Zotero operations, see
references/zotero-guide.md

2. Reading Modes

Mode	Trigger Phrases	Output
Quick Summary	"quickly look at", "quick"	3-5 sentences of core contributions
Full Analysis	"detailed analysis", default	Structured notes (using template)
Critical Analysis	"critically analyze", "critique"	Evaluation of methodological strengths and weaknesses
Knowledge Extraction	"extract formulas", "technical details"	Formulas + algorithm pseudocode

3. Note Generation

Template: Strictly follow

assets/paper-note-template.md

, no self-simplification allowed.

Core Quality Rules

No Omissions: All Figures, formulas, and Tables in the paper must be included in the notes
Inline Concept Links: Technical terms appearing for the first time in the text must be linked using
```
[[Concept]]
```
, not just at the end
No ASCII Flowcharts: Describe architectures using structured Markdown lists +
```
$mathematical symbols$
```
Formula Completeness: Each formula must have a name (
```
[[Concept|Name]]
```
), LaTeX formula, meaning, and symbol explanation
Priority to Image External Links: arXiv HTML / project homepage / GitHub; download locally only if not found

Detailed quality specifications for formulas/images/tables can be found in
references/quality-standards.md

Image Fetch Process (Multi-Source Fallback)

Goal: Ensure the notes include all Figures from the paper. First count the total number of Figures in the paper, then fetch them one by one.

WebSearch
```
"{paper title} arxiv"
```
to get the arXiv ID
Source A — arXiv HTML (preferred):
- WebFetch
```
https://arxiv.org/html/{arxiv_id}
```
  to extract titles and img src URLs of all
```
<figure>
```
  elements
- Count the total number of Figures in the paper and confirm if the extracted quantity is complete
Source B — Project Homepage (when HTML returns 404 or images are incomplete):
- Find the project homepage URL from the abstract/HTML (common patterns:
```
project page
```
  ,
```
github.io
```
  ,
```
our website
```
  )
- WebFetch the project homepage and extract displayed images (usually includes teaser/demo images)
Source C — PDF Extraction (when the first two sources fail):
- Use
```
pdfimages -png
```
  to extract from PDF, filter valid images larger than 10KB
Embed in notes using
```
![Figure X](url)
```
external links
Verification: External links are loadable / local files are >10KB
URL Deduplication: Check for duplicate arxiv_id path segments in the URL (e.g.,
```
2603.05312v1/2603.05312v1/
```
) before writing; remove duplicates if present. See
```
references/image-troubleshooting.md
```
for details.

ar5iv numbers do not necessarily correspond to Figure numbers; see
references/image-troubleshooting.md
for troubleshooting

Image Reliability Guarantee (Automatically Executed After Generation)

After saving the notes, run the image accessibility check script to automatically download inaccessible external link images to local:

bash

python3 ../daily-papers/download_note_images.py "{full note path}"

Accessible external links remain unchanged; inaccessible ones are automatically downloaded to
```
assets/
```
and replaced with Obsidian wikilinks
If localization is performed, the frontmatter
```
image_source
```
is automatically updated to
```
mixed
```

Formula Format

Each formula must include: name (

[[Concept|Name]]

), LaTeX

$$

block (with blank lines before and after), meaning, and symbol list. There must be blank lines before and after the

$$

block, otherwise Obsidian will not render it. Split extra-long formulas using

aligned

.

4. Obsidian Saving

File Naming

Use only the method/model name:

{method_name}.md

(e.g.,

Pi05.md

, no year prefix). Method name judgment: before the colon in the title / "We propose XXX" in the Abstract / convert Greek letters to ASCII. Save to

_To_Organize/

if unsure.

Save Path

Follow Zotero category hierarchy:

{NOTES_PATH}/{zotero_collection_path}/{method_name}.md

YAML Frontmatter

yaml

---
title: "Paper Title"
method_name: "MethodName"
authors: [Author1, Author2]
year: 2025
venue: arXiv
tags: [tag1, tag2]  # lowercase hyphenated, 3-8 tags
zotero_collection: 3-Robotics/1-VLX/VLA
image_source: online
created: YYYY-MM-DD
---

Tags judgment: check Related Work subheadings + Abstract keywords. The first tag is the core theme.

Automatic Execution After Saving

Refresh the directory pages only if

AUTO_REFRESH_INDEXES=true

:

bash

python3 ../_shared/generate_concept_mocs.py
python3 ../_shared/generate_paper_mocs.py

Perform git operations only if

GIT_COMMIT_ENABLED=true

:

First confirm that
```
VAULT_PATH/.git
```
exists
Ensure there are actual staged changes after
```
git add {new file} {paper_notes_folder}/
```
Execute the following only if conditions are met:

bash

cd {VAULT_PATH} && git add {new file} {paper_notes_folder}/ && git commit -m "add paper note: {method_name}"

Push only if
```
GIT_PUSH_ENABLED=true
```
and the remote repository is configured

5. Concept Library Maintenance (Required for Each Paper)

Concept library location:

{CONCEPTS_PATH}

Process

Scan all
```
[[Concept]]
```
links in the paper notes
Check if the corresponding concept notes exist for each link (using
```
ls
```
+
```
find
```
)
Create non-existent concepts (cannot be skipped), and automatically classify them into corresponding subdirectories

Classification rules and templates can be found in
references/concept-categories.md

Self-Check

Do all
```
[[Concept]]
```
links in the notes have corresponding concept notes?
Do the concept notes include this paper as a "representative work"?

6. Post-Completion Self-Check (Combined Checklist)

Are all Figures included in the notes (quantity matches the paper)?
Are all formulas included in the notes (variables consistent, no conflicts)?
Are all Tables completely retained (all rows and columns)?
Are technical terms in the text linked with
```
[[Concept]]
```
inline links?
Has the concept library been updated (missing concepts have been created)?
Are images usable (external links loadable / local files >10KB)?

7. Interactive Features

After completing the analysis, ask the user: Would you like an in-depth explanation? Compare with other papers? Save to Obsidian? After saving, automatically create missing concept notes and report the number of newly added concepts.

8. Batch Processing

Supports batch processing of Zotero categories (recursive subcategories by default). Process: recursively retrieve papers → deduplicate → skip existing notes → process sequentially → summarize.

Reference Files (Access As Needed)

references/zotero-guide.md
— Zotero query, classification, PDF path retrieval, intelligent category judgment
references/image-troubleshooting.md
— ar5iv image number correspondence, PDF extraction alternatives
references/concept-categories.md
— 16 subdirectory rules + templates for automatic concept classification
references/quality-standards.md
— Detailed quality specifications + self-checklists for formulas/images/tables