DeepPaperNote v2.0.0

DeepPaperNote v2.0.0 🚀

DeepPaperNote v2.0.0 is a major upgrade, and we strongly recommend that all users update to this version for a deeper, more reliable, and more Obsidian-native paper reading workflow.

What’s New ✨

1. Much Deeper Paper Notes 🧠

DeepPaperNote now produces notes that go beyond high-level summaries. The new workflow is designed to capture the full evidence chain of a paper: what the paper proves, what it does not prove, which experiments matter, how the results support the claims, and where the limitations change the interpretation.

The result is a note that feels closer to a real deep-reading research note, not just a compressed abstract.

2. Stronger Evidence-First Reading 🔎

v2.0.0 puts more emphasis on reading from the actual paper source. The workflow now builds a stronger source manifest and uses raw paper sections as the core reading material before drafting the final note.

If the available PDF or extracted evidence is too weak to support a real deep note, DeepPaperNote is expected to stop and ask for better source material instead of producing a low-confidence finished note.

3. Better Handling of Different Paper Types 📚

Different papers need different reading strategies. v2.0.0 improves support for multiple paper types, including method papers, benchmark papers, dataset papers, survey papers, and clinical or psychology empirical papers.

For example, benchmark and dataset papers now receive more focused treatment of data sources, task definitions, label design, evaluation protocols, privacy boundaries, and reproducibility limits.

4. More Insightful Analysis Sections 💡

The final note now aims to explain not only what the authors did, but why the design choices matter. DeepPaperNote places more emphasis on:

• central claims and their supporting evidence
• what the evidence actually proves
• what remains unproven
• negative or limiting results
• how mechanisms or protocols explain the result pattern
• comparison with baselines, prior routes, or human references
• reusable research and engineering takeaways

This makes the note more useful for future research, replication, and project planning.

5. Clearer Experiment and Result Presentation 📊

v2.0.0 improves how key numbers, comparisons, and ablations are presented. When a paper reports results across multiple models, datasets, tasks, settings, or metrics, DeepPaperNote is now more likely to organize the central comparison into compact Markdown tables followed by interpretation.

This makes result sections easier to scan and easier to reuse later.

6. More Reliable Figure and Table Handling 🖼️

Figure and table handling has been significantly improved. DeepPaperNote now makes a clearer distinction between three separate questions:

• Does the image match the intended paper figure or table?
• Is the crop visually usable?
• Should it be inserted into the final note?

When a figure or table candidate is usable and has a valid image path, it should be inserted as a real image. Placeholders are now reserved for real problems such as missing candidates, visual defects, contamination, truncation, identity mismatch, unresolved review, or copy/write failure.

This removes the old behavior where a clear usable figure could be left as a placeholder merely because it was considered lower priority.

7. Better Obsidian Output 🗂️

DeepPaperNote v2.0.0 is more Obsidian-native. Final notes now use a more consistent structure, including YAML properties, a fixed core metadata block, stable paper-local image folders, and cleaner figure/table embeds.

The workflow treats the configured Obsidian vault as the real save target. It should not silently fall back to a temporary workspace output and present that as a completed Obsidian save.

8. Smarter Paper Folder Routing 🧭

v2.0.0 improves how papers are routed into Obsidian folders. Instead of relying only on broad method categories, DeepPaperNote now uses an editable domain taxonomy and prioritizes application domains first.

This helps papers land in more meaningful folders, such as mental health, healthcare, software engineering, education, finance, robotics, or other application areas, before falling back to general method domains like machine learning or large language models.

9. Cleaner Chinese Writing Quality ✍️

The new version strengthens the language quality gate for Chinese notes. It is better at avoiding awkward mixed Chinese-English lines, mechanical translation artifacts, and half-translated technical phrases.

The goal is not just to pass formatting checks, but to produce notes that read naturally and are comfortable to keep in a long-term research vault.

10. A More Complete Final Review Process ✅

v2.0.0 adds a stronger final quality review before the final readability polish. The workflow now checks whether the note has enough analytical depth, whether key settings and numbers are covered, whether conclusions are properly bounded, and whether the note contains useful takeaways for future work.

This makes the final output more consistent and reduces the chance of shallow but well-formatted notes.

Project Homepage 🌐

https://917dhj.github.io/DeepPaperNote/

DeepPaperNote v1.1.1

Small update that fixes several existing logic gaps in DeepPaperNote.

Highlights:

• Tightened figure placeholder validation so retained placeholders must use the standard [!figure] callout format.
• Strengthened table crop quality checks to reject contaminated or mixed-caption crops.

Project homepage: https://917dhj.github.io/DeepPaperNote/

The attached DeepPaperNote.zip is a clean manually installable skill package. Unzip it and place the DeepPaperNote folder into your agent's skills directory.

DeepPaperNote v1.1.0

Minor release with a major figure/table extraction quality upgrade.

Highlights:

• Added figure-level PDF asset extraction that renders caption-anchored page regions instead of relying only on raw xref image objects.
• Improved extraction for complete figures, vector-heavy papers, fragmented LaTeX tables, and caption-on-bottom tables.
• Preserved DeepPaperNote's placeholder-first behavior: extracted figure assets are exposed as candidates, not automatic note insertions.
• Added visual quality signals so weak crops can fail closed and remain placeholders.
• Included figure_assets in the synthesis bundle so model-side review can inspect richer figure/table candidates.
• Added regression tests for figure asset candidates, placeholder-first planning, label normalization, and visual quality rejection.

Contributor note:

• This release incorporates the figure-level extraction work from PR #1 by KuangjuX, with follow-up changes to keep insertion semantics placeholder-first.

The attached DeepPaperNote.zip is a clean manually installable skill package. Unzip it and place the DeepPaperNote folder into your agent's skills directory.

DeepPaperNote v1.0.1

Patch release after v1.0.0.

Highlights:

• Added YAML frontmatter and wikilink rules for Obsidian-native features.
• Fixed lint_note.py compatibility with YAML frontmatter.
• Added tests for frontmatter stripping and frontmatter-aware lint compatibility.
• Fixed wikilink target resolution with a lookup-first, fail-closed approach.
• Removed unused image assets that were no longer referenced by the README files.

The attached DeepPaperNote.zip is a clean manually installable skill package. Unzip it and place the DeepPaperNote folder into your agent's skills directory.

DeepPaperNote v1.0.0

First stable release of DeepPaperNote.

DeepPaperNote is now a pure cross-agent skill for Claude Code, Codex, Cursor, Copilot, Gemini CLI, and other Agent Skills-compatible environments.

Highlights:

• The root SKILL.md is the single canonical skill entrypoint.
• Installation now supports npx skills add 917Dhj/DeepPaperNote -a codex and npx skills add 917Dhj/DeepPaperNote -a claude-code.
• Removed experimental onboarding/setup pseudo-surfaces and Claude plugin wrapper structure.
• Added AGENTS.md and CLAUDE.md for repo-level agent guidance.
• Preserved the evidence-first deep-reading pipeline, Obsidian-first output behavior, figure/table placeholder policy, lint gate, and final readability review.
• Added explicit Python >=3.10 interpreter guidance for agents running bundled scripts.

The attached DeepPaperNote.zip is a clean manually installable skill package. Unzip it and place the DeepPaperNote folder into your agent's skills directory.

DeepPaperNote v0.3.2-alpha

This alpha refresh strengthens local PDF metadata resolution so Zotero-style attachment names no longer dominate enrichment, while preserving the existing title, DOI, and arXiv workflows.

What’s New

• 📄 Local PDF metadata now prefers embedded PDF title, DOI, arXiv identifiers, and first-page title signals before falling back to cleaned filenames.
• 🧹 Local-PDF-only title correction can now replace noisy attachment-style titles with high-confidence external matches without changing the global merge policy.
• 🏷️ Candidate scoring now prefers published DOI and venue matches over preprint-style alternatives when both are available.
• 🔤 Common PDF ligatures such as fi and fl are normalized during extraction, so resolved titles are cleaner and more stable.
• 📦 The release package has been rebuilt from the latest main branch state for v0.3.2-alpha.

DeepPaperNote v0.3.1-alpha

This alpha refresh republishes DeepPaperNote on top of the latest main branch and aligns the default Obsidian paper output path with the current recommended vault layout.

What’s New

• 📂 The default Obsidian paper root is now Research/Papers instead of 20_Research/Papers.
• 🧭 Runtime path resolution and save behavior are aligned with the new default, so newly generated notes land in the updated location more consistently.
• 📦 The release package has been rebuilt from the latest main branch state for v0.3.1-alpha.

DeepPaperNote v0.3.0-alpha

This release is a substantial quality upgrade focused on making DeepPaperNote produce notes that feel more like durable, reusable deep-reading research notes rather than polished abstract summaries.

What’s New

• ✨ Stronger note structure: added and stabilized a dedicated Innovation section near the beginning of the note.
• 🧠 Clearer method understanding: method and system papers now more explicitly reconstruct the execution chain through a Mechanism Flow section.
• 📉 Better ablation coverage: notes are less likely to focus only on best-case results and now pay more attention to failed settings, weaker variants, and trade-offs.
• 🈶 Clearer abstract handling: the opening abstract block is now framed as Original Abstract Translation, making it much clearer that this section should be a Chinese translation of the paper’s original abstract rather than a newly written summary.
• 🧾 Cleaner metadata block: Core Info is now treated as a fixed metadata zone, so analysis or personal judgment is less likely to leak into it.
• 🚦 Stronger workflow discipline: the skill is now more explicit about following the full pipeline instead of silently skipping steps or downgrading behavior.
• 🔍 Added final readability review: after script lint passes, the model must reread the full note once more to improve fluency and remove awkward phrasing or unnecessary English leftovers.
• ∑ Safer formula handling: added a math syntax gate to catch common Obsidian / MathJax rendering failures before final save.
• 📂 More reliable saving behavior: fixed the occasional duplicated paper-slug directory issue and tightened Obsidian write / fallback rules.

DeepPaperNote v0.2.0-alpha

Highlights

• Stronger replication-oriented note writing for technical papers
• Explicit short note_plan before final note generation
• Equation-aware output when formulas are central to understanding the method
• Stricter final self-review for key numbers, method depth, and technical completeness
• Original abstract section now keeps both the English original and a Chinese translation
• Stronger formatting checks for suspicious mid-sentence line breaks and math accidentally rendered as code
• Chinese README is now the default GitHub homepage

What works now

• Single-paper deep-reading note generation for Codex
• Obsidian-native output with one folder per paper and paper-local images/
• Workspace fallback output when no Obsidian vault is configured
• Zotero-first helper workflow for local-library-first resolution
• OCR fallback for low-text PDF pages
• Placeholder-first figure handling with conservative image replacement
• Minimal test suite and GitHub Actions CI

Known limitations

• Chinese remains the only fully supported note language today
• High-confidence figure replacement still depends on extraction quality and semantic matching confidence
• Different sessions may still expose different python3 interpreters depending on environment inheritance

Recommended setup

• Configure your Obsidian vault for the best long-term note workflow
• Add a Zotero MCP option if you already manage papers in Zotero
• Install OCR dependencies if you often read scanned or low-text PDFs

DeepPaperNote v0.1.0-alpha

Highlights

• First public alpha release of DeepPaperNote as a Codex skill.
• Turns one research paper into a high-quality Markdown deep-reading note.
• Uses a model-first workflow: scripts gather evidence, the model does the paper understanding and final writing.
• Supports placeholder-first figure handling so missing images do not erase important note structure.

What Works Now

• Single-paper note generation from title, DOI, arXiv ID, URL, local PDF, or Zotero-resolved input.
• Obsidian-native output with folder-per-paper layout.
• Workspace fallback output when no Obsidian vault is configured.
• Zotero-first helper flow for local-library-first paper resolution.
• OCR fallback for low-text PDF pages.
• Domain-aware routing that prefers existing vault domains before creating a new one.
• Paper-local images/ folder creation during final save.
• Minimal automated tests and GitHub Actions CI.

Recommended Setup

• Configure an Obsidian vault for the best note-management experience.
• Optionally configure Zotero MCP if you already manage papers in Zotero.
• Optionally configure OCR tools for scanned or low-quality PDFs.
• Optionally configure a Semantic Scholar API key for stronger metadata fallback.
• Use /deeppapernote doctor or /deeppapernote start to inspect the local setup.

Known Limitations

• Figure replacement quality still depends on PDF extraction quality and semantic matching confidence.
• Some environments may expose different python3 interpreters across sessions; doctor now reports the active interpreter explicitly.
• Zotero integration quality depends on the user's available Codex-compatible Zotero MCP workflow.

Notes

This is an alpha release. The core workflow is usable, but the project is still being actively refined around figure handling, environment consistency, and broader real-world testing.