This repo gives builders a packaged skill for writing journal cover letters from manuscript files, then a trainer for improving that skill with blind drafts and permitted expert examples. The workflow separates Research, Review, and Bibliometrics letters so the argument matches the paper type.
Builders who want an agent to write or improve journal submission cover letters from manuscript materials.
You can draft a journal cover letter that matches the paper type and journal, then audit or improve it with a repeatable trainer workflow.
Uses separate Research, Review, and Bibliometrics workflows so each letter argues from the right kind of paper.
Reads the manuscript, title page, and supplements to build the letter from source material instead of guesswork.
Checks current journal guidance before writing and keeps the letter aligned with submission rules.
Returns an audit alongside the draft so the result can be reviewed for missing facts and mismatches.
Compares a blind AI draft with a permitted expert letter, extracts reusable reasoning, and turns it into a tested skill update.
Provides a `.skill`, a plugin `.zip`, and a portable `SKILL.md` for different agent setups.
Write a journal cover letter from manuscript files—or improve the Skill with manuscript and expert-letter examples.
| I want to... | Use | What it does |
|---|---|---|
| Write a cover letter | Journal Cover Letter Skill v3.2 | Reads the manuscript, checks the journal, writes the letter, and audits the result |
| Improve the Skill | Cover Letter Skill Trainer v0.2.0 | Compares a blind AI draft with a permitted expert letter and turns useful lessons into a tested Skill update |
| File | Choose it when... |
|---|---|
.skill | You want to install one Skill in Codex |
Plugin .zip | You want the complete Codex package |
SKILL.md | Your AI can read an instruction file but cannot install Codex packages |
The default output is English. You can request another language.
| Route | The letter should answer |
|---|---|
| Research | What scientific problem was studied, what was found, and why does the finding matter? |
| Review | What new understanding appears when the existing evidence is brought together? |
| Bibliometrics | What does the publication map reveal about the field's structure, growth, fragmentation, collaboration, or direction of travel? |
A bibliometric paper studies publications, citations, keywords, authors, institutions, and networks. Its main result is a map of a research field—not a treatment effect, mechanism, or ordinary summary of study findings.
A useful bibliometric cover letter asks:
The journal may label the paper Review or Original Research. The Skill keeps that official label while still using the Bibliometrics writing route.
The Trainer removes the need to repeat the improvement process by hand:
Current Skill
→ manuscript materials
→ blind AI cover letter
→ permitted expert cover letter
→ compare the two decisions
→ extract reusable reasoning
→ revise the Skill
→ test again in a fresh context
→ release only after review
The Trainer learns how the expert selected facts, ordered the argument, and made the work relevant to an editor. It does not copy the expert's wording. The manuscript remains the source of truth, because an expert letter can still contain an older title, number, journal detail, or declaration.
A clean first draft is not enough for a blind improvement loop. After the expert letter is revealed, every candidate must be generated in a new context that has not received the expert material. See the beginner Trainer guide.
An expert letter is a benchmark, not a gold standard. Persuasion never overrides factual accuracy or evidence boundaries. Private manuscripts and real letters stay outside this repository.
| Version | What the comparison revealed | What changed |
|---|---|---|
| v1.0 | A useful letter must first keep titles, numbers, authors, declarations, and journal rules accurate | Added a manuscript fact sheet, journal checks, permission for older letters, and a final audit |
| v2.0 | Research and Review should not use the same argument | Research tells one scientific discovery story; Review explains the new understanding created by synthesis |
| v2.1 | Review letters often report what was reviewed without saying what changed | Added field diagnosis, a memorable interpretation, measured promotion, and direct reader relevance |
| v2.2 | Research letters can let complex methods and result lists hide the discovery | Put the scientific finding first and describe methods by the credibility they add |
| v2.3 | Review conclusions could still be broad and the workflow could repeat questions already answered | Added a clear “old reading → synthesis finding → changed decision” contrast, a fast path for complete evidence, and stronger checks against empty promotion |
| v3.0 | Bibliometrics is neither ordinary Research nor a traditional Review | Added a third Bibliometrics route and separated the journal's official article label from the way the letter should argue |
| v3.1 | A polished bibliometric letter can become so abstract that it loses the paper's own results | Required a manuscript-specific map, frontier signal, or directional shift and added blind benchmark testing |
| v3.2 | Expert letters can contain stale facts, and bibliometric letters can overfocus on rankings or keywords | Keeps manuscript facts authoritative, joins performance analysis with science mapping, makes journal fit specific, and removes submission-system-only details |
Trainer development:
| Version | What changed |
|---|---|
| Trainer v0.1.0 | Added blind baseline generation, expert comparison, reusable rule extraction, candidate building, and regression checks |
| Trainer v0.2.0 | Requires a fresh expert-free context for every candidate round and records the real isolation level |
See the full development story and Trainer changes.
See the privacy policy.
Published papers and their cover letters can help improve this project. Share only material you have the right to disclose, and remove personal, confidential, and unpublished submission information first. See CONTRIBUTING.
Most users can ignore the folders below and download a Release directly.
| Path | Purpose |
|---|---|
skills/ | Current Writer and Trainer source |
docs/ | Chinese homepage, beginner guides, privacy, and version story |
development/ | Fictional examples, tests, and package builders |
.codex-plugin/ | Writer Plugin metadata |
.github/ | Automated checks and contribution forms |
Created by Jizhou Hu, China Medical University.
Licensed under the MIT License. Citation metadata is available in CITATION.cff.
Sign in to join the discussion.
No comments yet. Be the first to say what this is good for.
Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.
Topic in, narrated explainer video out. A Claude Code / Codex skill that turns any topic into a black-canvas motion-graphics explainer video with TTS voiceover, subtitles and a chapter progress bar. Chinese or English; every frame drawn in code with Remotion.
Public repository for Agent Skills
Open-source AI job search: scan job portals, evaluate listings into a structured A-H report with a global 1-5 score, tailor your CV, track applications — runs locally in your AI coding CLI (Claude Code, Codex, OpenCode, Antigravity…)
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
A skill to stop your coding agent from burying the answer. ADHD-friendly output.