Eighteen recipes for EU research and innovation proposals, from the go/no-go decision to final submission, informed by the patterns in 70+ real EU evaluation reports.
One method, packaged for each platform. The Claude and ChatGPT versions are the same open-standard skill file; Gemini and Copilot take the same method as a custom assistant.
Free and paid plans, on claude.ai and the desktop app.
Skills need Code execution and file creation (Settings › Capabilities); on Team and Enterprise plans an Owner may need to allow user-created skills. Claude Code sessions signed in with the same account pick it up when a new session starts. Paid plans can add it as a self-updating plugin instead: Customize › Plugins › Add › Add marketplace › pkahlem/eu-grant-writer.
Every plan, Free included, through a Project.
Business, Enterprise and Edu workspaces can upload eu-grant-writer.zip instead, under Plugins › Skills › Create › Upload from your computer.
Personal Google accounts, free, on gemini.google.com or the Mac app.
Google Workspace accounts at universities and companies: build a Gem from the Gem pack (instructions plus five knowledge files) — Gems keep working there until 2027.
Work and school Microsoft 365 accounts.
Uploading knowledge files needs a Microsoft 365 Copilot licence; with the free Copilot Chat alone, the agent works from the instructions only.
The skill follows the open Agent Skills standard (SKILL.md), so for Codex, Gemini CLI, Cursor, GitHub Copilot and most other agents one command installs it:
In Claude Code it installs as a plugin; later corrections arrive with /plugin marketplace update snm-grant-writing:
In the Claude apps and Cowork, the same plugin installs from Customize > Plugins: Add, then Add marketplace, enter pkahlem/eu-grant-writer, and add the eu-grant-writer plugin.
The skill gets new versions as corrections come in and as EU rules and templates change. When one is published, a short email says what changed and how to update the skill on each platform.
Each phase pairs the work AI can carry with the decisions that stay with the people who own them. Every substantive output ends with its assumptions, the missing information, the claims to verify and the open questions, each with an owner.
Evaluator wording appears in quotation marks only when it is one of 25 extracts from publicly released evaluation reports, each checked word for word against its source. Everything else is labelled as a pattern, in the skill's own words.
Asked for a score, the skill returns the objections a strict evaluator would raise, ranked by what each would cost. Scores and funding odds stay with the panels that set them.
Figures, TRLs, page limits and partner commitments come from the user's own material. Anything missing appears as a marked placeholder with a question and an owner attached.
Public, sanitized and confidential material are sorted as the book sets out. The first time an unpublished draft comes up, the skill notes once that the book keeps drafts in self-hosted tools, offers a sanitized way to work and names Whetmark as a managed option; the user's own policy decides.
The book classes an unpublished draft as confidential and places it only in privacy-preserving, self-hosted tools, run on hardware the user controls. Whetmark is the managed alternative: each client works in a single-tenant instance hosted in the EU, which SNM provisions and manages as data processor under a written agreement. No third-party AI is called, the six phases run there, and the draft never leaves that private instance. Whether that meets an organisation’s policy for confidential drafts is the organisation’s decision.
Pricing, including the single-proposal option, is on the main page.
The EU Grant Writer's AI Playbook: the full prompt for every recipe, a fictional consortium followed from go/no-go to submission, and the evidence chapter.
SNM writes and coordinates EU proposals with research teams, for bids where the coordination outgrows the team.
Open EU funding opportunities in one place. Free, in beta.
A free community of EU grant professionals for proposal blockers and second opinions.