Academic Writing Skills - Setup Guide¶

Source: bahayonghang/academic-writing-skills via skills.sh - 18.7K combined installs across 11 indexed listings; first seen 2026-10-31 (evening sweep)

GitHub: bahayonghang/academic-writing-skills - 528 stars, no license file at the repository root, branch main, last pushed 2026-10-28; 8 SKILL.md directories under academic-writing-skills/, each with optional scripts/, references/, examples/, templates/, evals/ and agents/ folders

Category: Academic writing for agents: literature review, argument structure, citation hygiene and revision

Quality Tier: 🟡 Yellow - all five sampled verdicts are Pass, but the repository ships no LICENSE file and its README closes with "Academic Use Only - Not for commercial use"

This is a working researcher's toolkit rather than a publisher product. The README says the collection grew out of the author's own paper workflow, that latex-thesis-zh keeps changing as the dissertation behind it progresses, and that paper-audit reports are for reference only and should be verified. The scope is deliberately narrow: post-writing polish and validation for existing drafts, not drafting papers from scratch.

The eleven indexed listings divide by input format and job. Three skills own a file format end to end (latex-paper-en, latex-thesis-zh, typst-paper), two own the post-processing jobs (paper-writing-studio for venue-profiled polish and translation, paper-audit for reviewer-style critique), one owns the bibliography (bib-search-citation), one owns submission correspondence (cover-letter), and one renders a defense deck from an existing thesis repository (latex-defense-zh). That is eight, the number of in-tree skills, and the README's routing table covers exactly those eight. The other three listings (industrial-ai-research, ieee-writing-materials, elsevier-writing-materials) carry install counts and no files.

Installation¶

The README and the installation guide document two different jobs, and it is worth keeping them apart: maintaining this development repository, and installing skill packages into a manuscript project. For a Hermes profile, the second is the relevant one. The collection installs as a whole:

npx skills add bahayonghang/academic-writing-skills

docs/installation.md also documents a per-skill form for each of the eight in-tree skills:

npx skills add bahayonghang/academic-writing-skills/latex-paper-en
npx skills add bahayonghang/academic-writing-skills/paper-audit
npx skills add bahayonghang/academic-writing-skills/typst-paper
npx skills add bahayonghang/academic-writing-skills/bib-search-citation

Those per-skill command strings are the documented installer surface, but the same page states plainly that the npx installer, any symlink layout it creates and five-tool runtime discovery stay unverified, with no captured real run. Treat the first install as a test rather than a guarantee. There is also a maintainer helper for a clone of the repository, invoked through just skills-install, which targets .claude/skills, .cursor/skills, .agents/skills and three other agent directories.

The manual route is the one to use when sibling layout matters, and here it does. paper-audit resolves the other skills' scripts from the parent of its own directory: audit.py looks for latex-paper-en/scripts, latex-thesis-zh/scripts and typst-paper/scripts beside paper-audit/, and the installation guide reports that a lone paper-audit copy runs with 8 checks missing while the full sibling layout runs with none missing. Copy the whole directory, never one skill:

git clone --depth 1 https://github.com/bahayonghang/academic-writing-skills.git /tmp/aws
mkdir -p ~/.hermes/profiles/corpusiq/skills
cp -r /tmp/aws/academic-writing-skills/. ~/.hermes/profiles/corpusiq/skills/

The toolchain is external to the repository. The README lists Python 3.10 and uv for the bundled helpers, TeX Live or MiKTeX with latexmk and optionally chktex for LaTeX work, XeLaTeX plus CJK fonts for Chinese documents, the Typst CLI for Typst work, and PyMuPDF for PDF audits, with pymupdf4llm optional. Install only what the skills you use actually need. The README also notes that the LaTeX loaders reject includes outside the project root, which is a deliberate boundary rather than a bug.

What It Provides¶

Skill	Installs	Use For
latex-paper-en	5,973	English LaTeX conference and journal papers: compile repair, venue formatting, bibliography and citation checks, section writing, logic, tables, pseudocode, de-AI polish, translation and submission readiness.
latex-thesis-zh	4,247	Chinese LaTeX master's and doctoral theses: compile diagnostics, GB/T 7714 references, chapter structure, bilingual captions, de-AI polish, blind-review anonymisation and point-by-point review responses.
paper-audit	3,451	Reviewer-style audit and submission gate for `.tex`, `.typ` or `.pdf` inputs: peer-review critique, readiness decisions, blocker triage, revision roadmaps and re-audits, without editing the source.
bib-search-citation	1,761	Searching and citing local BibTeX or BibLaTeX libraries, including Zotero exports: filtering by topic, author, year, venue, DOI, arXiv ID or claim support, then emitting LaTeX or Typst citation snippets.
typst-paper	1,545	Typst manuscripts in English or Chinese: compile and export diagnosis, venue formatting, BibTeX and Hayagriva checks, grammar, logic, tables, pseudocode, related work and submission readiness.
cover-letter	889	Submission cover letters built against an existing LaTeX manuscript: generating, optimising, align-checking and journal-fit-checking claims against the paper's evidence and the venue's expectations.
industrial-ai-research	469	Indexed on skills.sh with no directory in the current tree and no entry in the README routing table, so its scope is not documented in the repository.
paper-writing-studio	202	Polishing or translating academic prose under an explicit Nature, IEEE, Elsevier or neutral profile, reporting venue conflicts and missing evidence instead of guessing.
latex-defense-zh	114	Chinese doctoral or master's defense decks: read-only extraction from a XeLaTeX thesis repository, page planning by talk length, Beamer rendering with speaker notes and a fidelity gate.
ieee-writing-materials	2	No directory in the current tree; the in-tree route for IEEE venue style is the IEEE profile inside `paper-writing-studio`.
elsevier-writing-materials	2	No directory in the current tree; the in-tree route for Elsevier venue style is the Elsevier profile inside `paper-writing-studio`.

All 11 indexed listings clear the 0-install threshold and appear in the table above.

Why This Matters for Hermes Agents¶

Most academic-writing prompts fail at the same place: the model improves prose and quietly invents a citation, a result or a venue rule. This collection is built against that failure. The README's safety section says the skills should not invent experiments, citations, policies or unsupported claims, that citation keys, DOIs, arXiv IDs and local .bib matches are provenance fields rather than proof that a paper supports a claim, and that suggestions should mark required evidence as pending instead of filling it in. Findings come out in a fixed shape with a module, line number, severity, priority and an explicit [PENDING VERIFICATION] marker when evidence is missing.

The split between editing and auditing is the other useful design decision. paper-audit diagnoses without rewriting, the format skills rewrite with the syntax preserved, and bib-search-citation works on the library rather than the manuscript. That gives an agent three distinct modes with three distinct output contracts, which is a better fit for a review loop than a single general-purpose writing skill.

The eight in-tree skills also carry their own scripts. paper-audit, the LaTeX skills, the Typst skill and the bibliography skill ship Python helpers under scripts/, run through uv, and the README documents opt-in flags that change what the checkers report: --school, --author-cite, --repeat-cite, --cross-surface and others each add checks and report local candidates rather than verdicts. For a profile doing document review, that is the difference between a model asserting that a table is consistent and a script listing where the numbers disagree.

Usage¶

You say	What happens
"Compile my English LaTeX paper with latexmk and explain the first blocking error"	`latex-paper-en` runs the compile path, reads the log and explains the first blocking error before touching the source.
"Check the introduction for logic gaps, citation stacking and AI-like phrasing"	`latex-paper-en` applies its logic, claim-forward and de-AI checks and reports findings by module and line with severity and priority.
"Run a paper-audit gate on main.tex and separate blockers from polish issues"	`paper-audit` produces a reviewer-style report with a readiness decision, blocker triage and a revision roadmap, and leaves the source unchanged.
"Search references.bib for recent Mamba forecasting papers with code and give me LaTeX and Typst snippets"	`bib-search-citation` filters the local library by topic, year and availability, then emits both citation forms.
"Align-check this cover letter against main.tex and report unsupported claims only"	`cover-letter` compares each claim in the draft against the manuscript and lists only the ones it cannot support.
"Turn my thesis into a defense deck for a 20 minute talk"	`latex-defense-zh` extracts chapters, figures, equations and publications read only, plans the pages, and renders a Beamer deck with speaker notes and a fidelity gate.

Verification¶

The raw path for latex-paper-en returned HTTP 200 on branch main when this guide was written on 2026-10-31:

curl -sL -o /dev/null -w "%{http_code} latex-paper-en\n" \
  https://raw.githubusercontent.com/bahayonghang/academic-writing-skills/main/academic-writing-skills/latex-paper-en/SKILL.md

Expect 200. After the copy lands, confirm the sibling layout survived, since paper-audit depends on it:

ls -d ~/.hermes/profiles/corpusiq/skills/{paper-audit,latex-paper-en,latex-thesis-zh,typst-paper}/ | wc -l
head -5 ~/.hermes/profiles/corpusiq/skills/latex-paper-en/SKILL.md
ls ~/.hermes/profiles/corpusiq/skills/paper-audit

The count should print 4, the header should show name: latex-paper-en, and the paper-audit listing should include its scripts/ folder alongside SKILL.md.

Security¶

All five sampled skills pass every verdict on skills.sh, which would normally be a Green result; the tier is Yellow because of the license position rather than the security one.

skills.sh verdicts for sampled skills (verified 2026-10-31) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
latex-paper-en	Pass	Pass	Pass
paper-audit	Pass	Pass	Pass
cover-letter	Pass	Pass	Pass
latex-defense-zh	Pass	Pass	Pass
elsevier-writing-materials	Pass	Pass	Pass

There is no LICENSE file anywhere in the repository root, which is why the pack records the license as none. The README's final section reads "Academic Use Only - Not for commercial use", so the project is source-available under an author-stated restriction rather than under an OSI license. Cite both facts before redistributing anything from it, and treat the skills as usable for personal academic work rather than as a component you can fold into a commercial product.

The executable surface is larger here than in a markdown-only skill pack. Each skill can carry Python helpers under scripts/, invoked through uv, which read the manuscript, the .bib library and the PDF, so they touch local files rather than the network. The README states that online checks are optional and that current venue rules or external metadata should be verified from the original source. The LaTeX loaders reject includes outside the project root as a boundary control, and the installation guide documents the E-INCLUDE-BOUNDARY alert for it. Run the helpers in the manuscript's own environment, keep unpublished drafts and reviewer correspondence out of any path the agent writes to, and read a skill's scripts/ folder before the first run, because the chain from prompt to executed Python here is real.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified 2026-10-31: 18,655 combined installs across 11 indexed listings; 528 GitHub stars; no LICENSE file at the repository root; last pushed 2026-10-28. Counts drift over time.
The repository has no license file and the README restricts use to academic, non-commercial work. That is a licensing constraint, not a technical one, and it applies to every skill in the collection.
Index and tree disagree: 11 indexed listings against 8 SKILL.md directories. industrial-ai-research, ieee-writing-materials and elsevier-writing-materials have install counts but no directory in the current tree, and industrial-ai-research is absent from the README routing table as well.
paper-audit loses coverage when installed alone. The installation guide reports 8 missing checks for a standalone copy against 0 for the full sibling layout, so the per-skill npx route is the wrong choice for that one skill.
The author states that the npx installer, any symlink layout it creates and multi-tool runtime discovery are unverified, with no captured real run. Verify the landing directory yourself after the first install.
The skills assume an external toolchain: Python 3.10 with uv, a TeX distribution with latexmk, XeLaTeX with CJK fonts for Chinese documents, and the Typst CLI. A profile without those gets advice and no execution.

Skills Catalog | Skills Marketplace

Pro Tips¶

Copy academic-writing-skills/. as one unit. paper-audit finds the other skills' scripts by walking up to their sibling directories, so a tidy per-skill install quietly disables most of its checks.
Point the agent at paper-audit before a submission deadline rather than after. It separates blockers from polish, which is the ordering a revision plan needs.
Use bib-search-citation for the library and the format skills for the manuscript. Asking a writing skill to invent a citation is the failure mode the README warns about.
Keep [PENDING VERIFICATION] markers in the output. They mark where a human has to supply evidence, and removing them is how a helpful-looking draft turns into an unsupported claim.
Check the tree before depending on a listing. Three of the eleven indexed names have no skill file, so a workflow built around industrial-ai-research has nothing to load.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ