Skip to content
Get packs

Guide · Agent skill

How to use the Scientific Writing Assistant agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The Scientific Writing Assistant is an agent skill that helps an AI agent draft, revise and audit scientific manuscripts or reports with traceable evidence. This guide explains what the skill changes in the agent's behaviour, how to install it, how to trigger it, and what to watch for on confidentiality and scripts. By the end you will be able to install the skill, ask for a checked draft section, and know which checks remain your responsibility.

Scientific Writing AssistantK-Dense-AI/scientific-agent-skills · MIT

Quick start

  1. 1Download the ZIP. Download the scientific-writing ZIP from this page.
  2. 2Unzip the skill folder. Unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder.
  3. 3Or upload it in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP.
  4. 4Describe your task. Ask in plain language for a specific section, check or audit, naming the study design.
  5. 5Provide your verified sources. Share your source manifest and claim IDs so the agent binds claims to evidence rather than inventing them.
  6. 6Run optional local checks. If you want, run the bundled offline scripts with Python 3.11+ to check consistency and claims.
  7. 7Verify before submitting. Have a human check every source, metadata field and journal requirement before any submission.

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder with a SKILL.md file of instructions that an AI agent can load when a task matches the skill's description. This skill, named scientific-writing, is for manuscript sections, references, declarations, tables, figures and submission preparation where scientific accuracy and traceability matter.

Once installed, the agent separates drafting from verifying. It links each factual or numeric claim to recorded evidence IDs, points to a reporting guideline such as CONSORT or PRISMA for the study design, and flags missing or inconsistent details instead of filling them in. It also reminds you about confidentiality, authorship accountability and AI disclosure.

  • Refuses to invent citations, DOIs, results, sample sizes, approvals or funding statements.
  • Uses explicit missing or unverified states instead of plausible boilerplate.
  • Keeps methods and results consistent and names any analysis-set differences.
  • Warns that AI is not an author and that generated fluency is not evidence.

Install the skill from the ZIP

Download the ZIP for skills/scientific-writing from this page. Then choose one of two routes.

For a personal install, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. Alternatively, in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

  • Personal: ~/.claude/skills/
  • Project only: .claude/skills/ inside your project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check what the skill needs before you run it

The skill's written guidance is platform-neutral. Its optional local command-line scripts need Python 3.11 or later and no extra dependencies. The skill does not use the network, and the bundled tools are offline and need no API keys.

The scripts are optional. You can use the skill for drafting and checklists without running them.

  • Python 3.11+ (only for the optional local scripts)
  • Runs scripts: yes, locally
  • Uses network: no
  • No API keys required

How do you trigger the skill with a natural request?

You do not need a special command. Describe the scientific writing task in plain language and name the document part you need. The agent loads the skill when the request matches its description.

Good requests state the document type, the study design, and what you want checked. Give the agent the verified sources and claim IDs you already have, and say clearly if a fact is not yet verified.

  • "Draft the Methods section of my randomized trial report using only the evidence IDs in my source manifest."
  • "Which reporting guideline fits a systematic review of diagnostic accuracy studies?"
  • "Check whether the sample sizes in my Results table match the Methods text."
  • "Prepare my manuscript for submission and list what is still unverified."

Run the optional local checks

The skill's workflow includes local scripts. For a new draft, it can generate Markdown, JSON and CSV scaffolds in a workspace folder. The generator never overwrites files, and its output is explicitly not submission-ready. Other scripts check consistency between repeated numeric facts, validate the source manifest, audit claims against the manuscript, and check references.

The scripts are run from the skill's folder with python3. Check references/cli_reference.md for the full set of options before running them on your own files.

  • scripts/scaffold_manuscript.py creates a draft workspace.
  • scripts/select_reporting_guidelines.py helps choose a guideline by study design (non-scoring).
  • scripts/check_consistency.py checks a consistency manifest.
  • scripts/validate_manifest.py, scripts/audit_claims.py and scripts/check_references.py verify sources and claims.
python3 scripts/scaffold_manuscript.py --output-dir ./draft-workspace --document-id local-draft --study-design randomized_trial --guideline consort-2025

Keep confidentiality and verification under your control

Do not send unpublished manuscripts, peer-review material, restricted data or patient data to an external service without explicit authorization and a documented policy review. When authorization is unclear, the skill keeps processing local. Removing obvious names does not count as de-identification, which needs expert review.

A human must open each source, confirm the claim and locator, verify the bibliographic metadata, and record who verified it and when. Search snippets and generated summaries do not verify a claim.

  • Never treat the agent's output as verified references or results.
  • Check the target journal's current author instructions yourself.
  • Make the final authorship and AI-disclosure decisions yourself.

What people use it for

  • Drafting a Methods section where each design detail is tied to a recorded source.
  • Choosing the reporting guideline for a trial, systematic review or diagnostic accuracy study.
  • Checking that sample sizes and denominators match between the Results text and tables.
  • Preparing a thesis chapter's references and claim list for supervisor review.
  • Auditing a draft for missing declarations such as funding, ethics or data availability.

Frequently asked questions

Is the scientific-writing skill free to use?

The skill is released under the MIT licence, as stated on this page. The bundled tools run offline and need no API keys, and the source does not list any price.

What do I need to install the scientific writing skill?

You can install it by unzipping the downloaded folder into ~/.claude/skills/ or a project's .claude/skills/ folder, or by uploading the ZIP under Settings, Capabilities, Skills in the Claude apps. Python 3.11+ is needed only for the optional local scripts.

Does the skill write references or DOIs for me?

No. The skill forbids inventing citations, references, DOIs, PMIDs or URLs. It cannot verify literature itself, so a human must open and check every source.

Does the skill send my manuscript to the internet?

The skill's stated design is offline, and the source says it does not use the network. It still warns you not to send unpublished or restricted material to any external AI service without institutional authorization.

Which reporting guideline does the skill recommend?

It chooses by actual study design and article type, such as CONSORT 2025 for randomized trials or PRISMA 2020 for systematic reviews. Its selector is non-scoring and does not certify compliance, so you must check the current official guideline yourself.

How does this compare with simply asking an AI to write my paper?

This skill separates drafting from verification, binds claims to evidence IDs, and flags gaps rather than filling them. A general prompt does not enforce these checks, so the skill is useful when traceability matters.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are a researcher, thesis writer or research team who wants an AI agent to draft and audit manuscripts with traceable claims, reporting-guideline checks and consistency reviews. Pick something else if you need the agent to find or verify literature for you, or if you need expert review of de-identification, ethics or journal policy compliance.

Download the skill