How to use the Grant Proposal Writer agent skill

What the grant-proposal skill makes the agent do

The skill follows a multi-step workflow to turn research ideas into a grant application for a funding agency you choose. It surveys the literature and checks novelty, designs the proposal structure (aims and narrative), drafts the prose in the agency's expected format and language, and then runs an external-style review and revision loop.

Grant proposals argue for future work (feasibility and potential), not completed results. The skill is built around that distinction. It can also optionally mirror the layout of a past proposal or paper you point it to.

• Literature survey and novelty check before drafting
• Proposal structure design, including aims and narrative
• Prose drafted in the agency's expected format and language
• Review and revision loop, with output saved as GRANT_PROPOSAL.md

Install the grant-proposal skill

Download the ZIP for the skill from the page where you found it. The skill's files include SKILL.md. Install it in one of two ways.

For Claude Code, unzip the download into ~/.claude/skills/ to make it available for all your projects, or into a project's .claude/skills/ folder to make it available only in that project.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need to remember a command. The skill activates when you ask for a grant proposal or funding application. The source lists these phrases as triggers: 'write grant', 'grant proposal', 'write KAKENHI', 'NSF proposal', and the Japanese phrases 申請書, 科研費, 基金申请 and 写基金. It also activates when you say you want to turn research ideas into a funding application.

You can name the agency and sub-type in the same request. The skill's own examples use the form below, with the research direction before the dash and the grant type after it.

• 'Write a KAKENHI proposal on my work on low-power sensor networks'
• 'Grant proposal for an NSF CAREER application on protein folding simulation'
• '基金申请 for NSFC Youth on my catalyst research'
• 'Turn these validated ideas into an ERC application'

Choose the grant type and options

The default grant type is KAKENHI. Other supported types are NSF, NSFC, ERC, DFG, SNSF, ARC, NWO and GENERIC. Name the type in your request to override the default. Sub-types are detected from your request, or the most common one is used. Examples in the source include KAKENHI Start-up, Wakate and Kiban-B, NSFC Youth, Excellent-Youth, Distinguished, Overseas and Key, and NSF CAREER, CRII and Standard.

Output is Markdown by default. You can ask for LaTeX. The language is set automatically from the grant type (for example Japanese for KAKENHI and Chinese for NSFC), and you can override it in your request.

The optional style reference (— style-ref: followed by a source) is off unless you add it. It uses a helper script, tools/extract_paper_style.py, to read the structural layout of a past paper or proposal. Sources can be a local TeX folder or file, a local PDF, an arXiv id, or an http(s) URL. Overleaf URLs and IDs are rejected, so clone the project locally first.

• Default grant type: KAKENHI; override by naming another agency
• Default output: Markdown; LaTeX available on request
• Language follows the grant type unless you override it
• Style reference is opt-in and only mirrors structure, not wording

What you need before you start

The skill needs Claude Code or another agent environment that is compatible with SKILL.md. It also needs web search and fetch access for literature lookup. The source describes these as required.

Some features are optional. An external reviewer through Codex MCP or another reviewer MCP server can be used for external review rounds, depending on the skill's reviewer configuration. The style reference needs the helper script described above.

• Required: Claude Code or a SKILL.md-compatible agent environment
• Required: web search and fetch access for literature lookup
• Optional: Codex MCP or another reviewer MCP server for external review rounds
• Optional: tools/extract_paper_style.py for the --style-ref option

Safety notes: scripts, network and sign-off

The skill runs scripts and uses the network. Its tool permissions include shell commands, file reading and writing, editing, search and web access. Review what it runs if this matters to your environment.

The skill is designed to stop at each checkpoint and wait for your explicit confirmation. Its automatic-proceed setting is off by default. Grant content needs expert judgment and accuracy checks, so you should review every section yourself before submission.

The skill relies on its own knowledge and web lookups for agency rules, which may be outdated. Check current formatting rules with the agency. Give it real results only. Because it argues for future work, weak inputs can lead it to make unsupported claims.

• It runs scripts and uses the network
• It waits for your confirmation at each checkpoint by default
• Agency formatting rules may be out of date; verify them with the agency
• Do not submit without PI review and accuracy checks

Tips for better drafts

Start with validated research ideas and a clear feasibility argument. The skill organises ideas into a fundable narrative, so vague inputs produce vague drafts. Describe planned work rather than results you have not obtained.

Name the agency and sub-type early. The structure, sections and language depend on it. Use the review loop output as a list of questions for your own revision, not as a finished text.

• Bring validated ideas and a feasibility argument
• Describe future work, not results you do not yet have
• State the agency and sub-type in your first request
• Treat reviewer feedback as a checklist for your own revision