Skip to main content
Guide9 min read·Updated July 12, 2026
🧩

Best AI Agent Skills for Grant Writing in 2026

B

A. Frans

Published July 12, 2026

AI Agent SkillsGrant WritingClaude CodeNonprofitsFundraising

Grant writing is 20% persuasion and 80% logistics: decoding a 40-page RFP, matching your program to a funder's priorities, building a budget that survives scrutiny, and formatting the whole thing to a submission spec that rejects you for a wrong margin. AI agent skills won't write a proposal that wins on merit your program doesn't have. What they do is collapse the logistics (the reading, the reformatting, the budget math, the compliance checks) so the human hours go into the case for support, which is the only part a reviewer actually scores.

A skill, if you're new to the term, is a folder of instructions and scripts that teaches an AI agent like Claude Code to do a specific job well. You install it once and the agent knows how to build a Word document, read a PDF, or calculate a budget without you re-explaining it every session. These are the ones worth installing for grant work in 2026. If your organization does this daily, this pairs with our list of AI tools for writers.

The skills at a glance

SkillWhat it does for grantsInstalls asSource
PDF ToolsReads RFPs, extracts requirements, fills form-based applicationsSkillanthropics/skills
DOCX (Word)Produces submission-ready proposals with correct formattingSkillanthropics/skills
XLSX (Spreadsheet)Builds and checks budgets, budget justificationsSkillanthropics/skills
Deep ResearchFinds funders, reads their priorities and past granteesSkillcommunity repo
Doc Co-authoringDrafts and revises narrative sections with youSkillcommunity repo
The first three come from Anthropic's official skills repository, which is the cleanest install story in this whole category. The last two are community skills that solve real grant-writing problems but need the security check every third-party skill deserves. More on that below.

PDF Tools: turning the RFP into a checklist

Every grant starts with a document nobody wants to read closely and everybody gets dinged for skimming. The RFP or funding announcement buries the disqualifiers (eligibility rules, page limits, required attachments, the exact evaluation criteria) across dozens of pages, and missing one is how good programs get tossed before review.

The PDF skill lets the agent read the whole thing and pull it apart. Point it at the RFP and ask for every hard requirement, every deadline, every attachment, and the scoring rubric, and it returns a structured checklist you can actually work against. Then, when the application is a fillable PDF form (still depressingly common with government funders), the same skill fills it programmatically from your answers instead of you tabbing through fields by hand.

Install it from Anthropic's repository:

git clone https://github.com/anthropics/skills.git

Copy the pdf folder into your skills directory:

~/.claude/skills/pdf/SKILL.md

Once it's there, Claude Code loads it automatically when a task involves a PDF. The payoff is turning a 40-page announcement into a one-page compliance checklist in a couple of minutes, which is the difference between catching the "no indirect costs over 10%" line now versus after you've built the budget around 15%.

DOCX: the formatting tax, paid automatically

Funders reject on format. Wrong font, wrong margins, a page over the limit, a missing section header they specified: reviewers use these as the first cut, because they get more applications than they can read. Getting the document exactly right is tedious and it's the last thing you want to be doing at 11pm on deadline day.

The DOCX skill hands that off. It generates real Word documents (proper headings, tables, page numbers, styles) from your content, matched to the funder's spec. Tell the agent the required structure (statement of need, project design, evaluation, budget narrative) and the formatting rules, and it assembles a submission-ready file. It reads existing Word docs too, so it can take last year's funded proposal and restructure it for a new funder's template.

Same install source, the docx folder:

git clone https://github.com/anthropics/skills.git
~/.claude/skills/docx/SKILL.md

Where this earns its place is version three through nine of a proposal. Every revision that used to mean manually reflowing a document, checking you're still under the page limit, and re-applying styles becomes a single instruction. You spend the reclaimed time on the argument, not the layout.

XLSX: budgets that add up

A grant budget is where amateur applications fall apart. The numbers don't reconcile with the narrative, the indirect-cost rate is applied wrong, the personnel lines don't sum, or the justification contradicts the spreadsheet. Reviewers notice, and it reads as sloppiness about money, the worst signal you can send someone about to give you money.

The XLSX skill builds and audits the budget. It creates spreadsheets with working formulas, so when you change a salary line the totals and the indirect calculation update correctly. It cross-checks the budget against the narrative for consistency, catches the arithmetic errors that creep in during last-minute edits, and produces the budget-justification breakdown many funders demand. For a multi-year grant with matching requirements, that math-checking alone is worth the install.

From the same repository, the xlsx folder:

~/.claude/skills/xlsx/SKILL.md

Because these three (PDF, DOCX, XLSX) all live in one repo, a single clone gets your whole document toolchain. That's the strongest reason to start here: one trusted source, one command, three of the four grant-writing jobs covered.

Deep Research: finding the right funder

The best-written proposal sent to the wrong funder loses. Prospect research, finding funders whose priorities, geography, and typical grant size match your program, and reading who they've funded before, is slow, and it's the work that most improves your odds because it happens before you write a word.

A deep-research skill turns the agent loose on this. It searches for funders in your area, reads their guidelines and 990 filings, summarizes their recent grantees, and flags where your program fits and where it doesn't. Instead of a week of manual foundation-directory digging, you get a shortlist with reasoning attached. It's also useful mid-proposal for pulling the statistics and evidence your statement of need has to cite.

This is a community skill rather than an official one, so treat the install with care. Find it on GitHub, read the SKILL.md and any scripts it ships before you add it, and only install from a repo you can inspect. The security note in the next section applies most to skills like this.

Doc Co-authoring: the narrative, with a second brain

The parts a reviewer scores (the need statement, the project design, the theory of change) still have to be written by someone who understands the program. A co-authoring skill doesn't replace that writer; it works alongside them, drafting sections from your notes, tightening a bloated paragraph, checking the narrative answers every point in the RFP's rubric, and keeping terminology consistent across a 15-page document.

Used well, it's the editor a small nonprofit can't afford to keep on staff. Used lazily (accept the AI draft, submit it), it produces exactly the generic, evidence-thin proposal reviewers have learned to score low. The skill is a drafting partner, not a ghostwriter, and the applications that benefit are the ones where a human drives every claim.

Like Deep Research, this is a community skill. Audit before you install.

The security note you shouldn't skip

A skill is code and instructions the agent will run with your permissions. An official skill from anthropics/skills is a known quantity. A skill from a random GitHub account is not, and grant work means the agent is touching sensitive documents: budgets, board information, sometimes beneficiary data.

Before installing any third-party skill: open the SKILL.md and read what it tells the agent to do. Check any bundled scripts for network calls or file access that doesn't fit the stated job. Prefer skills with real stars, recent commits, and a readable author. Install from a fork you've reviewed, not a URL you were handed. This takes ten minutes and it's the cheapest insurance in your workflow. The official document skills earn their front-of-list spot partly by removing this question entirely.

Putting it together on a real deadline

A funded workflow looks like this. Deep Research finds and vets the funder and pulls your evidence. PDF Tools reads the RFP into a compliance checklist and, if needed, fills the application form. Doc Co-authoring drafts the narrative sections from your program notes, which you then rewrite in your own voice with your own data. XLSX builds and audits the budget against the narrative. DOCX assembles the final submission-ready document to the funder's exact spec, and you run the checklist one last time before you hit submit.

The agent carried the logistics; you carried the case. That's the right split, and it's what lets a two-person development shop compete with organizations that have a full grants department.

FAQ

Can an AI agent write a winning grant proposal on its own? No. It can handle research, formatting, budget math, and first drafts, but a winning proposal rests on a strong program and specific evidence only your team has. Reviewers score merit and fit, and unedited AI narratives read as generic. Use skills for the logistics and drive the argument yourself.

Which skills should I install first? Start with the three official ones, PDF, DOCX, and XLSX, because a single clone of anthropics/skills gets all three from a trusted source. They cover reading RFPs, formatting proposals, and building budgets, which is most of the grunt work. Add research and co-authoring skills after, with a security review.

Are agent skills the same as ChatGPT or a plugin? Not quite. A skill is a folder of instructions and scripts that teaches an agent like Claude Code to do a specific task, loaded automatically when relevant. It's more capable than a chatbot prompt because it can run scripts (build a real Word file, calculate a budget), not just generate text.

Is it safe to install community grant-writing skills? It can be, with review. Read the SKILL.md and any scripts before installing, check the repo's activity and author, and install from a source you can inspect. Grant work involves sensitive financial and program data, so an unaudited third-party skill is a real risk. Official skills from anthropics/skills remove that concern.

Do I need to be technical to use these? Basic comfort with a terminal helps for the one-time install (a git clone and copying a folder). After that, you work in plain language, "read this RFP and list every requirement", and the agent handles the rest. A nonprofit staffer who can follow install instructions once can run the whole workflow.

Share this article

📬

Get More AI Tool Guides

New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.