DeepPaperNote
AI agent skill that reads academic papers and generates structured Obsidian notes.
917Dhj · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add 917Dhj/DeepPaperNote技能说明
name: deeppapernote description: Generate a high-quality deep-reading note for a single paper and write it into an Obsidian-style vault. Use when the user gives a paper title, DOI, URL, arXiv ID, Zotero item, or local PDF and wants a polished Markdown note with strong structure, evidence-based analysis, and figure placeholders.
DeepPaperNote
Use this skill when the user wants one outcome:
- read one paper carefully
- generate a high-quality Markdown note
- save the note to the workspace or Obsidian target selected by resolved configuration
Chinese trigger examples:
给这篇论文生成深度笔记写一篇高质量论文精读笔记把这篇文章整理成 obsidian 笔记读这篇论文并生成 md 笔记
English trigger examples:
Generate a deep-reading note for this paperTurn this paper into an Obsidian research note
User Configuration
Before a normal paper run, read references/user-configuration.md for configuration admission, migration, repair, Run Overrides, and Preference Changes.
Resolve Run Overrides from the explicit request, CLI, and current process environment first. When they form a complete valid configuration for the selected Save Mode, Configuration Readiness is complete without reading User Configuration. Only inspect User Configuration when those Run Overrides need fallback values.
Language Integrity Contract
After Configuration Readiness, resolve one output_language (zh-CN or en) for the run. source_manifest.language_hint describes source text only and never selects the note profile.
Bind that exact value through Save Target Admission → Figure Plan → Figure/Table Decisions → Synthesis Bundle → Note Plan → Grounding Lint → Final Note Lint → Final Quality Review → Final Readability Review → Formal Save:
- Every JSON artifact in the chain carries a top-level
output_language; the Synthesis Bundle also carries the same value atwriting_contract.language. - Before producing its output, every adjacent consumer requires each input language and compares it with the resolved value. Missing, unsupported, or mismatched values stop the run; no stage infers or defaults an artifact language.
- Final Quality Review and Final Readability Review each receive the resolved value and check the note against only that profile.
- Final Note Lint records
note_sha256. Any review edit invalidates the prior lint, so rerun Final Note Lint under the same language. Formal Save requires the lint language andnote_sha256to match the final note, and validates Figure/Table Decisions language before any save side effect.
This contract is complete only when every named stage is bound to the resolved value and Formal Save validates the final bytes. Read references/output-language.md for profile content while drafting or debugging either language.
This skill is intentionally narrow:
- it is one canonical Skill and one pipeline with two output profiles
- it handles one paper at a time
- it does not update daily reading lists
- it does not treat a shallow abstract rewrite as a successful output
- it does not split the public entrypoint into separate setup, troubleshooting, or start commands
Core Standard
The finished note must be more than a summary. It should reconstruct the paper's argument:
- what problem it solves
- how the task is defined
- what data or materials it uses
- how the method or analysis actually works
- what results matter most
- what the paper does not prove
- why the paper is worth keeping
Default writer persona:
- a top-tier researcher or algorithm engineer
- writing a replication-oriented lab note
- not writing a popular-science explanation
- assuming the reader can follow Python, PyTorch, training loops, and evaluation logic
The note must adapt to the paper type. Use the same base structure, but shift emphasis for AI methods, benchmarks, clinical studies, and humanities or social-science papers.
Workflow
Follow this order:
- complete Configuration Readiness: resolve Run Overrides first, and inspect User Configuration only when they are incomplete; advance only after the resolved run configuration is complete and valid
- resolve the paper identity
- collect metadata
- acquire the best available PDF
- extract canonical raw source text:
*_raw_sections.jsonl,*_source_manifest.json, and optional derived*_full_text.md - perform Save Target Admission before drafting or domain routing:
- for Obsidian mode, run
scripts/write_obsidian_note.py --preflightwith the resolved title, exactoutput_language, Vault, and*_source_manifest.json; this program result is authoritative, so do not replace it with prompt-only duplicate checking - when admission returns
reuse_source_directoryorreuse_empty_same_name_directory, use that directory and skip domain selection - when it returns
same_language_note_exists, stop before drafting and ask whether to overwrite the reported note. If the user approves, rerun preflight with--overwrite-existing-note --expected-existing-note-sha256 <reported_sha256>and carry that exact confirmation into Formal Save; if the user declines, stop without writing - for any other blocked conflict, report the returned ambiguity and stop without creating a second directory
- workspace mode does not scan an Obsidian Vault and continues through its normal domain routing
- for Obsidian mode, run
- extract structural indexes and PDF assets
- plan figure placement
- build the full figure/table decision table
- build the manifest synthesis bundle
- have the model read the bundle plus raw sections and create a short JSON
note_planthat satisfies the generated bundle contract, including its exactoutput_language - draft from the plan only after the grounding gate passes
- have the model write the note
- lint the final note against the same
note_plan— this stage completes only when the lint artifact exists and every reportedpasses_*gate istrue; otherwise revise and rerun lint. If the lint output containspasses_style_gate: false, apply the Styl