research-paper-lifecycle-skills

Agent skills for the whole research-paper lifecycle: discover, write, verify, submit, rebut, publish, and present with rules checked from live sources.
Field-agnostic, citation-aware, and designed as a copilot: you stay the author, the agent does the careful checking.

_{Quick start · How it works · All 42 skills · Workflows · Integrations · Website · LLMs.txt · Boundaries}

Why this exists

Most AI writing tools stop at the draft. Papers get rejected after that: a page
over the limit, an author name left in a double-blind PDF, a missing checklist,
a hallucinated citation, a broken artifact, or a camera-ready step nobody had
time to read.

research-paper-lifecycle-skills gives compatible agents a structured
playbook for the research work around the paper. It can search and verify
sources, tighten writing without changing claims, check venue rules against
live CFPs or author instructions, triage reviews, prepare artifacts, and turn a
finished paper into slides, scripts, Q&A practice, or a poster.

Project website: https://shaishavmaisuria.github.io/research-paper-lifecycle-skills/
AI-readable summary: https://shaishavmaisuria.github.io/research-paper-lifecycle-skills/llms.txt

Who it is for: researchers, PhD students, labs, reviewers, and builders
supporting publication workflows in any discipline with public venue guidance.
The examples are academic-paper shaped, but the checks are intentionally
general: sourced rules, verified citations, local outputs, and human sign-off.

Quick start (60 seconds)

# Any Agent Skills-compatible environment
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills

# See what is included
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills --list

# Install one skill
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills --skill preflight-check

# Install into a supported agent target
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills --agent codex
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills --agent cursor
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills --agent gemini
npx skills add ShaishavMaisuria/research-paper-lifecycle-skills --agent claude-code

Claude Code plugin marketplace:

/plugin marketplace add ShaishavMaisuria/research-paper-lifecycle-skills
/plugin install paper-submission@research-paper-lifecycle-skills

Then ask in plain language:

Will this get desk-rejected at an IEEE conference? Check my main.tex against this CFP.

The right skill should trigger itself, load its SKILL.md, run any relevant
helpers, and return a sourced report.

No paper yet? Start with any paper folder

You do not need a bundled template. Put your draft files in a normal project
folder and let the skills build the research workflow around it.

1. Create the basics: main.tex or a manuscript draft, plus refs.bib if
   you already have references.
2. Set the context: ask for paper-profile, then say the target venue,
   field, contribution type, and deadline if you know them.
3. Run the first checks: ask "verify refs.bib", "find missing related
   work", "write the abstract for SIGSPATIAL", or "preflight my paper".

How it works

1. Skills: one folder per task. Each SKILL.md tells the agent when to
   use it, what to do, what to verify, and what not to touch.
2. Live rules: venue-aware skills build or re-check requirements from the
   live CFP, author instructions, or public policy page. Unsourced facts stay
   marked as unverified.
3. Deterministic helpers: bundled scripts handle repeatable checks such as
   citation resolution, anonymization linting, page-budget maps, timeline math,
   and dashboard rendering.
4. Local workspace: generated reports go into paper-workspace/; paper
   preferences and lessons can live in .paper-memory/. Both stay with the
   user's paper, not in this package.

Find the skill you need

Open the linked SKILL.md for trigger rules, process, guardrails, references,
and bundled scripts. The complete public package currently contains 42 skills.

Start here

Complete linked index

Common workflows

New paper:        paper-profile -> literature-review -> write-abstract -> draft-related-work -> polish-prose
Survey draft:     find-papers -> draft-survey -> verify-citations -> polish-prose
Before submit:    parse-cfp -> tailor-to-venue -> anonymize-paper -> preflight-check
Quality pass:     verify-citations -> check-originality -> benchmark-paper -> simulate-reviewers -> assess-paper
Reflection:       reflect-paper -> polish-prose / verify-claims -> reflect-and-improve
Length pass:      fit-page-limit -> polish-prose / refactor-structure / polish-tables-figures
Overleaf:         work-with-overleaf -> preflight-check / verify-citations / polish-prose -> work-with-overleaf
Artifacts:        test-research-code -> verify-results -> refactor-research-code -> prepare-artifacts
Reviews are in:   triage-reviews -> write-rebuttal
Accepted:         prepare-camera-ready -> make-slides -> write-talk-script -> rehearse-qa -> make-poster
End to end:       orchestrate-paper

Usage examples

Worked example: idea to a clean SIGSPATIAL submission

I have a method and results for streaming-trajectory indexing.
Get it submission-ready for SIGSPATIAL.

1. paper-profile records contribution type, target audience, constraints,
   and risk appetite.
2. find-papers, literature-review, and verify-citations build a verified
   reference base.
3. write-abstract, draft-related-work, polish-prose, and match-style
   shape the draft without changing claims.
4. parse-cfp, tailor-to-venue, anonymize-paper, and preflight-check
   verify live requirements and desk-reject risks.
5. fit-page-limit, simulate-reviewers, benchmark-paper, and
   assess-paper produce the review-readiness pass.
6. render-workspace-html gives a local dashboard of the resulting
   paper-workspace/ outputs.

Integrations

This is an Agent Skills package, not a single-agent prompt bundle. The
repository ships portable SKILL.md folders plus optional scripts and
references, so any agent that understands the Agent Skills format can use it.

Optional Claude Code plugin bundles currently exposed in
.claude-plugin/marketplace.json: paper-search, paper-writing,
paper-submission, and paper-presenting. Use the Agent Skills CLI above to
install the complete 42-skill package.

Manual install for tools with their own skills directory:

mkdir -p "$YOUR_AGENT_SKILLS_DIR"
SKILL_NAME=preflight-check
cp -R "skills/$SKILL_NAME" "$YOUR_AGENT_SKILLS_DIR/"

Memory and dashboards

Most tools forget you when the chat ends. This package uses a local
.paper-memory/ convention, described in
paper-memory-convention.md,
so skills can reuse paper positioning, writing preferences, decisions, and
lessons without uploading them.

reflect-paper can run paired Researcher and Writer reflection passes: one
checks evidence, novelty, positioning, and venue fit; the other checks story,
structure, tone, and reader friction. reflect-and-improve then asks whether a
change measurably helped before accepting it.

render-workspace-html turns paper-workspace/ into a local browser dashboard
with progress, artifact cards, and recent activity. The dashboard is
self-contained HTML, refreshable on request, and never uploaded by the skill.

What it won't do

• No acceptance predictions. Scores measure conformance, completeness, and
  quality signals; they do not predict reviewer decisions.
• No automatic submission. Skills can prepare, check, and explain, but they
  do not upload or submit a paper for you.
• No invented citations or venue facts. Unresolved citations and unsourced
  rules are marked as risks instead of being smoothed over.
• No bundled paper content. Paper PDFs and copyrighted corpora are not
  redistributed in this repository.
• No authorship laundering. The skills can polish language and surface
  issues, but you remain responsible for the work and final text.

Principles

1. Human-led. Skills draft, check, and explain; the author decides.
2. Citation-aware. Writing workflows route new references through
   verification instead of trusting plausible text.
3. Copyright-conscious. Paper content is fetched only from legal open
   sources, on demand, and is not bundled in this repository.
4. Venue-skeptical. Profiles and CFP facts can go stale; final rules must
   be checked live, with source links, before relying on them.
5. Portable. The repository ships skills, scripts, README visuals, and
   package metadata; not paper PDFs or cached research corpora.

Make it yours

The lifecycle is field-agnostic. To adapt it for another discipline, copy a
skill folder, adjust its SKILL.md triggers and guardrails, and keep the same
verification-first habit: sourced venue rules, verified citations, and local
outputs only.

Repository layout

assets/                         README and community visuals
docs/                           GitHub Pages landing page and crawl files
skills/                         42 agent skills
  <skill>/SKILL.md               instructions and guardrails
  <skill>/references/            supporting guidance
  <skill>/scripts/               deterministic helper scripts
.claude-plugin/marketplace.json  optional Claude Code plugin bundle metadata
.github/ISSUE_TEMPLATE/          issue forms for bugs, docs, and skill requests
.github/workflows/              public package validation workflow
CODE_OF_CONDUCT.md               community behavior expectations
CONTRIBUTING.md                  contribution guide
SECURITY.md                      vulnerability reporting policy
LICENSE                          Apache-2.0 license
NOTICE                           attribution notice
CITATION.cff                     citation metadata

Attribution and license

Copyright 2026 Shaishav Maisuria.

Licensed under the Apache License, Version 2.0. See LICENSE.

If you redistribute this package or include substantial portions of these
skills in another repository, keep the copyright/license notices and the
NOTICE attribution file so the original work is credited.

Disclaimer: venue rules change every year. Always confirm against the live CFP
or author instructions before submission.