Founder of the Year

A Claude Code plugin

The stage intro the skill made for its worked example: an arena announcer, a generated track, the founder's own sites, and his AI-rendered portrait at the name. 49 seconds. Made with the open-source /brag skill by Shunit Haviv Hakimi (MIT) and Hyperframes, with an ElevenLabs announcer voice and a track generated for the piece. The portrait is AI-rendered from the founder's own photograph. Founder-prepared, drafted with AI assistance.

Founder of the Year

A workflow that writes a confident, magazine-style profile of a founder and their businesses, as a web page, a PDF and Markdown, with a verification sheet behind it. It reads the founder's own sources first, asks one short batch of questions, checks every claim, then writes.

Founder-prepared and AI-assisted. The name is a creative brief, not a status: the output never implies an award, a publisher byline, an endorsement or independent verification, and says so on every page it produces.

See the feature it introduces Install

How it works

Six steps, each leaving a file you can read. Ask for one step or the whole run.

  1. Set upWho you are, who the reader is, who may be named and what stays private. Earlier drafts and interviews are imported, not redone.
  2. Gather firstIt finds out what it can reach, then reads your notes vault, CRM, code host, case studies and public records before asking you anything. Read-only, no credentials.
  3. Ask onceOne short batch of questions, each with a sensible default, for what the records cannot say. Interview questions only for the human story.
  4. Choose the storyA few candidate angles judged on evidence, interest and risk, then one is chosen.
  5. Write itHeadline, standfirst, at-a-glance panel, highlight band, case studies, timeline, exact pull-quotes. Delivered as a designed page, a PDF and Markdown.
  6. Check and hand overOne evidence check and one editorial read, then a few plain lines telling you what was made and where.

Honest by construction

  • A claim ledger. Each claim has a source, a date, a status (corroborated, attributed, conflicting, unverified) and a treatment. A source proves only what it can: code is not deployment, a CV is not independent corroboration, a company's own testimonial is the company's voice.
  • Public and private are different folders. The shareable pack and the private research record never mix, and superseded drafts go to a legacy folder instead of lingering.
  • Your words, not the model's. Anything attributed to you has to be in the interview record. Quotations are exact or they are paraphrase; none is reconstructed.
  • It audits you first. Before it flatters you it checks your own websites and the public registers, because that is where an editor looks first.
  • Caveats live in the sheet. The article reads like a profile, not an audit. Everything a sceptical editor would ask sits in a separate verification sheet.
  • Nothing leaves your machine on its own. It drafts a pitch; it does not send one. Publishing, posting and submitting are always your decision.

Four agents, narrow jobs

Researcher

Investigates one bounded question from primary sources and writes a dated source register.

Evidence reviewer

Tests every consequential sentence against the raw sources. Read-only.

Editorial reviewer

Reads the draft cold, as a stranger would, and judges the promise, the structure and the tone. Read-only.

Pack reviewer

Checks every document against the others for drift, overreach and privacy leaks. Read-only.

The process is modelled on published newsroom principles of accuracy, verification, transparency and accountability, for example the SPJ Code of Ethics and Reuters' standards. It is not affiliated with, endorsed by or checked by either, and AI review is not external verification.

A worked example, with a disclosure

The first real run was on the author's own story, so the example is a founder-prepared feature about the person who wrote this tool: a database specialist who took a startup's platform from voice notes to launch-ready and is building a software factory for small firms. It comes with its verification sheet, which shows which statements are public records, company pages, the founder's account or a design, and what is still unchecked. Read the sheet alongside the feature: that is the point.

Open the example

Install

In Claude Code:

/plugin marketplace add DEX-CON-ONE/founder-of-the-year
/plugin install founder-of-the-year@founder-of-the-year

Or, for other Agent Skills tools:

npx skills add DEX-CON-ONE/founder-of-the-year

Then ask, in plain words:

Use founder-of-the-year to write my founder profile. My notes are in my
vault and my client work is on GitHub; read those first and ask me only
what you can't find.

The plugin is plain Markdown plus one HTML template: a skill, four agents and ten reference files. It contains no code, hooks or servers; the PDF is printed with a browser you already have. Agents can use your agent's web tools when you authorise delegation, and your prompts go to your model provider under its terms.

What it is not

  • Not an award, a ranking or a publication, and not independent journalism.
  • Not a verifier: it checks claims against the sources you supply and public records, and says what it could not check.
  • Not a mass-pitch tool. The pitch is a draft for a human to rewrite, and many journalists object to AI-written pitches.
  • Not finished. It has been run end to end once, on a real founder, and improved from what that run found. Issues and criticism are welcome.