PaperStyle

中文说明

PaperStyle is a reusable repository for top-journal writing-style distillation and agent skill delivery.

The project is not limited to the current nature-family-optics sample room. The intended workflow is:

1. define a target constraint set such as a publisher family, journal group, topic area, or author corpus
2. retrieve and organize a representative empirical corpus
3. combine empirical examples with normative sources such as templates and formatting guides
4. distill an executable style profile
5. ship the result as a Codex-native skill plus validation and drafting utilities

The current public sample room is Nature family optics.

Release Boundary

The public repository ships:

• reusable scripts, configs, and skill definitions
• distilled profile artifacts
• lightweight placeholder documentation under data/targets/
• the mirrored Nature formatting guide used by the sample room

The public repository does not rely on bundling third-party publisher template packages or downloaded raw paper source archives. Those assets are expected to be provisioned locally under data/targets/<target_id>/... by the user or by an agent acting on the user's behalf.

See docs/target-asset-setup.md for the concrete directory contract, publisher template links, and Nature-family sample-room bootstrap steps.

Repository Layout

• configs/target_specs/: what we want to distill
• configs/corpus_queries/: how we search for empirical candidates
• configs/corpus_recipes/: how we assemble empirical and normative sources into a corpus
• configs/style_profiles/: executable distilled style profiles
• codes/: reusable project scripts for retrieval, corpus building, profile bootstrapping, drafting, and validation
• skills/: deliverable Codex skills
• data/targets/: target-scoped input assets and raw source data
• artifacts/targets/: generated manifests, corpora, packs, reports, and adapters
• docs/: public architecture and workflow documentation

Current Sample Room

The bundled sample room targets:

• publisher family: Springer Nature / Nature family
• domain: optics, photonics
• empirical corpus: 20 arXiv-source-accessible published papers
• normative corpus: 1 official LaTeX template expected locally + 1 tracked formatting guide mirror

Primary entry points:

• Architecture: docs/architecture.md
• Nature skill guide: skills/nature-family-optics/README.md
• Skill definition: skills/nature-family-optics/SKILL.md

Quick Start

Bootstrap the target-local assets first:

1. follow docs/target-asset-setup.md
2. place the target template package under data/targets/<target_id>/normative/
3. download or regenerate raw source material under data/targets/<target_id>/raw_sources/

Fetch or refresh the sample-room discovery report:

python codes/fetch_corpus_candidates.py

Download arXiv source packages:

python codes/download_arxiv_sources.py

Build the section-level corpus:

python codes/build_section_corpus.py

Bootstrap the style profile and adapters:

python codes/bootstrap_style_profile.py
python codes/distill_corpus_writing_patterns.py
python codes/render_adapters.py

Build and validate a manuscript pack:

python codes/build_manuscript_pack.py
python codes/validate_manuscript_pack.py

Normalize an external draft into the active target template:

python codes/normalize_manuscript.py --input /path/to/draft.md --compile-check

Release Notes

• Project-local absolute paths have been removed from public-facing configs and generated artifacts.
• Public docs live outside notes/.
• Target-scoped input data and generated artifacts are now separated into data/targets/ and artifacts/targets/.
• Third-party template bundles and raw paper source archives are now treated as local target assets rather than public repository payload.
• The repository keeps the codes/ layout for now; a future phase can promote these scripts into a package and CLI layer.