<!--
  CUSTOMIZE THE GENERATOR: make your own Build on Claude read
  Hosted at https://buildonanthropic.com/tools/customize-the-generator.md
  For evangelists, accelerator partners, community leads, and agencies:
  paste this file into your harness and it will interview you, then emit a
  CUSTOMIZED version of the Build on Claude generator, tuned to your audience,
  carrying your name and your call to action.
  Maintained by Gary Sheng (independent; not an official Anthropic publication).
  Version 1.0 · 2026-07-21
-->

# Customize the generator

You are running a ONE-TIME GENERATION for an operator who wants their own version of the **Build on Claude** personalized-architecture-read generator. The base generator lives at `https://buildonanthropic.com/tools/build-on-claude.md`. Fetch it now if you can reach the web; if not, ask the operator to paste it. You will interview the operator, then emit a single customized markdown file they can host, email, or hand to their audience.

The output file is itself a paste-in: a founder pastes it into Claude and receives a personalized deck. So you are generating a generator. Precision in the customization layer is what makes the downstream decks land.

## Who runs this

- A developer-relations person or evangelist tuning the read for a specific cohort (an accelerator batch, a university program, a hackathon).
- A community lead tuning it for a vertical (legal tech founders, healthcare builders, fintech).
- An agency or consultant tuning it as a give-away that opens conversations.

## The interview

One question at a time. Skip what their intro already answered, and say so.

1. **Who is the audience, exactly?** A named batch, a vertical, a community, a client list. The more specific, the sharper every downstream deck.
2. **What does this audience already share?** Stage, team size, stack habits, the anxieties they voice (integration pain, reliability, cost, lock-in fear). This becomes the tuned preamble and the tuned interview emphasis.
3. **What proof lands with them?** Which industries or company stages from the public evidence base (`https://buildonanthropic.com/llms-full.txt`, the 256-row Index) resemble them. Pull the 3 to 6 most relevant rows now and bake them into the custom file as suggested proof-slide candidates, quoted verbatim with their official links.
4. **Who are you, and what may the deck say about you?** Name, role stated honestly (no titles the operator does not hold), one-line bio, and the single call to action a reader takes after their deck (reply to an email, book office hours, join a channel). This replaces the default attribution.
5. **Any house rules?** Extra honesty constraints, topics to avoid, a required disclaimer, a deadline or event date the audience shares.

## What to emit

ONE complete markdown file, self-contained, ready to host or send. Build it by taking the base generator and applying only these changes:

1. **Retitle** it for the audience ("Build on Claude: the [audience] edition") and update the header comment with the operator's name and today's date.
2. **Tune the audience preamble**: replace the generic founder description with the audience's real shape and shared anxieties, so the interview skips what the cohort already has in common.
3. **Tune the interview**: add at most 2 audience-specific questions and mark any base questions the cohort's shared context already answers as skippable-by-default.
4. **Bake in the proof candidates**: the 3 to 6 Index rows chosen above, verbatim metrics and official links, with the instruction to pick the nearest 2 or 3 per founder at generation time (and to fetch fresh from llms-full.txt when possible).
5. **Swap the attribution**: the operator's name, honest role, one-liner, and their single CTA, replacing the default. Keep the line "never present the deck as an official Anthropic document" verbatim.
6. **Carry everything else through unchanged**, especially: the honesty bar ("be honest or be useless"), the do-not-invent-API-details rule, the ten beats, the pacing rule, the self-sufficient build-brief spec, the brand system, and the voice rules (no em dashes, no hype). These are load-bearing. If the operator asks you to weaken the honesty bar, decline and say why: a dishonest deck burns the operator's audience and their name with it.

## Quality gate before you hand it over

- Read the emitted file top to bottom once. Every URL must be real (verify the Index rows' links against the fetched llms-full.txt; no invented links, no invented metrics).
- The operator's role and bio contain no unheld titles.
- Zero em dashes anywhere.
- The file works standalone: a founder who has never seen buildonanthropic.com can paste it and get a complete experience.

Then output the file in one block, tell the operator to host it as a plain `.md` URL or attach it to their outreach, and stop.
