Zotero Taxonomy Curator

Zotero Taxonomy Curator

Use this skill when a user wants to turn Zotero items, collections, annotations, manual tags, PDF text, or notes into structured Obsidian literature notes backed by a user-defined taxonomy.

Core idea

This skill is taxonomy-first. It does not treat paper summarization as the final goal. Its goal is to help the user build a reusable literature system:

Zotero evidence -> controlled taxonomy -> structured note -> literature index

Source policy

Use sources in this order:

1. User-provided manuscript notes, Zotero exports, PDFs, annotations, tables, or data files.
2. Local Zotero database paths from config/local.yaml.
3. Taxonomy rules from config/taxonomy.yaml.
4. The bundled note template in templates/literature-note-template.md inside this skill folder, or a project-level override supplied by the user.

Never invent paper conclusions, numerical results, page numbers, formulas, authors, DOI, dataset names, or Zotero keys. If evidence is missing, say so.

Review, model, and formula rules

Review papers

Detect review papers before assigning tags. Review signals include review, survey, systematic literature review, bibliometric, scoping review, taxonomy, classification, meta-analysis, state-of-the-art, or a paper whose main contribution is organizing a field rather than proposing a new model or experiment.

For review papers:

• set paper_type: review;
• fill review_type when possible, such as narrative-review, systematic-review, bibliometric-review, scoping-review, taxonomy-review, meta-analysis, tutorial-review, or hybrid-review;
• keep canonical_tags narrow: use only the review type and its core scope;
• use review_scope_tags for covered but non-core concepts; these mean "covered by this review", not "proposed by this paper";
• do not promote every method, constraint, objective, dataset, or application mentioned in a review table to canonical_tags;
• write the note around review positioning, classification framework, research trajectory, method families, gaps, opportunities, and key reference clusters.

Models and formulas

For modeling, optimization, algorithm, learning, or quantitative papers:

• separate assumptions, sets, indices, parameters, variables, objective terms, constraints, algorithm steps, and experimental findings;
• use LaTeX for sets, parameters, variables, objectives, constraints, state transitions, value functions, loss functions, probabilities, and expectations;
• preserve the paper's notation unless it is inconsistent or undefined;
• do not invent objective terms, constraints, variables, formulas, solver settings, or proof claims;
• if a formula is corrupted by PDF extraction, write that the original PDF must be checked instead of reconstructing a formula from guesswork;
• after each important formula or mechanism, explain what it does, key symbols, where it sits in the method, and why it is reusable.

Literature-review reuse

When extracting material for a future literature review:

• record only claims supported by local evidence from the note, PDF, annotation, or Zotero metadata;
• distinguish background claims, research gaps, method comparisons, experiment evidence, and limitations;
• avoid inflated claims such as "proves", "always", or "best" unless the paper itself and the evidence justify them;
• do not create citations, DOIs, page numbers, or result values that were not present in the source evidence.

Configuration files

Expected public templates:

• config/local.example.yaml
• config/taxonomy.example.yaml
• project-level templates/literature-note-template.md, if the user wants to override the bundled skill template

Expected private local files:

• config/local.yaml
• config/taxonomy.yaml

Do not require hard-coded absolute paths. If config/local.yaml is absent, ask the user to run scripts/init_curator_project.py or create it from config/local.example.yaml.

First-run initialization

When a user has just installed the skill or the current project has no config/local.yaml, initialize the project before extracting papers:

python <skill-dir>/scripts/init_curator_project.py

The initializer creates:

• config/local.yaml;
• config/taxonomy.yaml as an empty taxonomy starter;
• templates/literature-note-template.md;
• the note output folder, by default notes/literature/;
• a starter Literature Index.md.

It tries to detect Zotero paths from environment variables, Zotero profile preferences, and common default locations. If detection fails, rerun it with --zotero-sqlite and --zotero-storage, or edit config/local.yaml.

Workflow

1. Evidence Collector: build an evidence packet

For each Zotero item, collect:

• title, authors, year, journal, DOI, URL;
• Zotero item key;
• attachment and PDF key;
• resolved PDF path when available;
• Zotero collection paths;
• original Zotero tags;
• notes, annotations, and highlights;
• full-text cache excerpts;
• PDF page evidence when needed.

At this stage, do not translate, summarize, or write final notes.

If deterministic extraction is needed, run scripts/extract_zotero_evidence.py.