Academic Research Skills for Claude Code

• Plan a paper chapter by chapter with Socratic questions

  Claude asks you questions and helps you decide each chapter before any drafting starts.

  Try: “Guide me through writing a paper on demographic decline”

  Write · /ars-plan

• Run the full research-to-final-paper pipeline

  Runs the whole pipeline: research, write, integrity check, review, revise and finalize, stopping for your approval at each stage.

  Try: “I want to write a research paper on AI's impact on higher education QA”

  Write · /ars-full

• Make a detailed outline and evidence map

  Builds an outline plus a map of which evidence supports which section.

  Write · /ars-outline

• Write an annotated bibliography

  Produces an annotated bibliography formatted as part of a paper.

  Write · /ars-lit-review

• Compare papers by WHY, HOW and WHAT

  Compares several papers side by side on why they were written, how they work and what they find.

  Write · /ars-3w

• Get a simulated peer review panel

  Runs a mock review panel on your draft and returns their comments.

  Try: “Review this paper”

  Write · /ars-reviewer

• Revise a draft and answer reviewers

  Produces a revised draft and point-by-point responses to reviewer comments.

  Write · /ars-revision

• Turn reviewer comments into a plan

  Reads reviewer comments and gives you a ranked roadmap for the revision.

  Write · /ars-revision-coach

• Check a rebuttal against the reviews

  Checks your rebuttal draft against the original reviews and flags points you missed.

  Read · /ars-rebuttal-audit

• Write a bilingual abstract and keywords

  Drafts an abstract in two languages with keywords.

  Write · /ars-abstract

• Convert to LaTeX, DOCX, PDF or Markdown

  Converts the paper between formats. DOCX needs Pandoc and PDF needs tectonic.

  Runs code · /ars-format-convert

• Get a citation error report

  Checks each reference against public indexes and reports the ones that look wrong or unsupported.

  Try: “Check citations”

  Network · /ars-citation-check

• Draft an AI-use disclosure for a venue

  Writes an AI-usage statement that fits a specific journal or conference.

  Write · /ars-disclosure

• Record that you read a cited work

  Marks a cited work as read by you, so the record shows which sources a person checked.

  Write · /ars-mark-read

• Withdraw a read record

  Removes your read mark from a cited work.

  Write · /ars-unmark-read

• Drop cached results for a citation

  Clears saved verification results for one citation so it is checked again.

  Delete · /ars-cache-invalidate

• Research team: sources, synthesis, systematic review

  Runs a team of research roles that frame the question, search the literature, check sources and synthesize findings. Modes include quick, lit-review, fact-check, socratic and systematic-review with optional meta-analysis.

  Try: “Do a systematic review on X with PRISMA”

  Write · deep-research

• Plan, draft and revise the paper

  Plans, drafts and revises a paper. Modes include outline-only, abstract-only, citation-check, disclosure and rebuttal-audit.

  Write · academic-paper

• Five-seat simulated peer review panel

  Simulates a journal-fit reviewer, three reviewers and a devil's advocate. Live reviews are labelled NOT_CALIBRATED.

  Write · academic-paper-reviewer

• Chain research, writing and review with checkpoints

  Orchestrates the other skills in 10 stages, with mandatory integrity checks before and after review and a process summary at the end.

  Write · academic-pipeline

• Screen studies for systematic reviews

  Screens studies with two blinded AI reviewers and an adjudicator. Reads RIS, PubMed .nbib, Web of Science and CSV, and outputs PRISMA 2020 counts and RIS groups.

  Try: “Screen these database exports”

  Write · sr-screener

• Used by the assistant behind the scenes

  Subagent: merges findings across sources

  Merges findings across sources, resolves conflicts and maps gaps.

  Read · synthesis_agent

• Subagent: designs the methodology

  Designs the methodology blueprint for a study.

  Read · research_architect_agent

• Subagent: compiles APA 7.0 reports

  Turns findings into an APA 7.0 report.

  Write · report_compiler_agent

• Subagent: blinded screener for sr-screener

  One of the blinded reviewers used by sr-screener.

  Read · screening_reviewer_agent

• Lists /ars-* commands and checks for updates

  Runs announce-ars-loaded.sh at session start. It adds a note listing the commands and at most once a day fetches plugin.json from GitHub. ARS_UPDATE_CHECK=0 turns this off.

  Runs code · SessionStart hook

• Local guard that blocks out-of-scope ARS writes

  Runs run_guard.sh before each write, edit or shell call. It finds Python and runs a guard that denies out-of-scope writes and all Bash for fenced ARS subagents. No network; passes through if Python is missing.

  Runs code · PreToolUse hook (Write, Edit, MultiEdit, Bash)

Watch out for

• A hook runs before every write and shell call

  The PreToolUse hook runs hooks/run_guard.sh before every Write, Edit, MultiEdit and Bash call in any session where the plugin is on. It starts a local Python guard that blocks out-of-scope writes by ARS subagents and makes no network calls. github.com

• Opt-in cross-model checks can send your manuscript out

  With ARS_CROSS_MODEL set and your consent each session, integrity samples or the full paper can go to OpenAI or Google using your own keys. It is off by default, so leave it unset for unpublished work. github.com

• The write guard has blocked edits outside ARS work

  Older versions blocked CLAUDE.md edits in every project (the author says fixed in v3.13.0). In Copilot's cloud agent the guard still refuses every edit, and on Windows it can slow down tool calls. github.com

• A start-of-session hook checks for updates

  The SessionStart hook runs scripts/announce-ars-loaded.sh, which lists the /ars-* commands and, at most once a day, fetches plugin.json from GitHub to compare versions. No user data is sent. Set ARS_UPDATE_CHECK=0 to turn it off. github.com

• Reference titles and IDs go to four public indexes

  Citation checks send DOIs, arXiv IDs and reference titles to Semantic Scholar, OpenAlex, Crossref and arXiv. If your reference list is confidential, skip the verification scripts. github.com

• Your drafts and instructions are saved to local files

  The Material Passport and run ledger record your instructions and checkpoint answers next to the path you choose. Integrity checks can save full source texts there, so keep them out of shared folders. github.com

Security advisories

• CClaude DesktopSupported
• CCodexSupported
• CClaude CodeHas issues
• CCopilot CLINot supported1 open issue
• GGemini CLINot supported
• OOpenCodeNot supported

Tap an app to see how to set it up and any known problems.

In their words