Two Claude skills that turn a sermon transcript into a printed small group guide. The first writes the guide; the second renders it as a 2-page branded PDF for your congregation.
A two-stage pipeline: paste the transcript, get the guide, ask for the PDF.
Why this exists
Most AI-generated sermon notes flatten the sermon. They read like a theological summary of what the pastor talked about rather than a guide that helps a small group re-encounter the actual sermon. The pastor’s anecdote about his neighbor, the half-quoted line from a theologian, the joke that landed, the moment the room got quiet: these get smoothed away by default, even though they are exactly what people remember mid-week and exactly what makes a sermon this sermon.
This bundle is tuned to do the opposite. It anchors itself in the actual flow and texture of the preaching, then derives the theology and discussion questions from there. Personality moments, named non-Scripture sources, and Scripture cross-references are each preserved differently, so a leader can scan the guide and find what they need.
The other half of the problem is presentation. A guide that looks like a wall of plain text gets skimmed, not used. The second skill renders the guide as a printable handout: logo at the top, a single accent color from your church’s brand, your congregation’s name in the footer. It looks like something the church made, because it is.
Why two skills
The split is deliberate. The first skill makes editorial decisions: what to keep, how to phrase the questions, how to preserve the preacher’s voice. The second skill makes design decisions: what colors, what typography, how the page breaks. They use different muscles, and forcing them into one skill made each worse. As separate skills, you can iterate on a guide’s content without re-rendering, or change the design without re-running the editorial work. The Markdown is the durable artifact in between, and you can also use it on its own (paste it into an email, post it on the website, hand it to a different renderer later).
What it produces
The first skill, small-group-guide, produces a Markdown guide with four sections:
- Big idea (1-2 sentences)
- Theological anchor points (2-4 concrete claims)
- Discussion questions (3-6, sermon-specific, no filler)
- Sermon walk-through (4-8 short paragraphs tracing the actual sermon arc, with inline italic callouts for personality moments and named sources, and parenthetical Scripture citations)
The walk-through distinguishes three categories of moment, and treats each one differently. Personality moments and named non-Scripture quotes get italic callouts; Scripture cross-references get parenthetical citations in the prose. Mixing them visually creates noise; separating them creates signal.
The second skill, church-pdf-render, takes that Markdown and renders it as a 2-page PDF: logo at the top, copper section headings (or your accent color), warm-cream big-idea callout, drop cap on the walk-through, italic personality callouts in warm brown, page footer with church name and page number, centered colophon at the end.
The renderer auto-tightens the layout once if a guide overflows to a third page. Two pages is the design target.
Who it’s for
Small group, community group, life group, or home group leaders who need a guide to facilitate discussion during the week. The Markdown output is designed to be scanned on a phone before group starts; the PDF is designed to be printed and handed out. Use one, the other, or both.
How to use it
- Drop both skill folders into your Claude skills directory (or upload them through claude.ai → Settings → Capabilities → Skills).
- Configure each skill once before first use. See “Configuring the skills” below.
- Paste a sermon transcript into a Claude conversation. The first skill should trigger automatically; if not, ask Claude to “use the small-group-guide skill.”
- When the guide looks right, ask for the PDF. “Make the PDF” or “render the guide” should trigger the second skill.
The transcript can be rough (auto-generated, no punctuation, missing speaker labels, even auto-translated from another language). The first skill is built to work with messy input.
Configuring the skills
Both skills are independent — you configure each one separately, and you can use either without the other.
small-group-guide reads church-context.md from its own folder. Open the file in a text editor and fill in the prompts: church name, tradition, primary teaching pastor, congregational vocabulary, theological frame. The model uses these to calibrate voice and theological register. Without this calibration, the output drifts toward a generic register that could come from any tradition.
church-pdf-render reads church-config.yaml from its own folder. Set the church name, location, and accent color. Drop your church’s logo at assets/logo.png. Other values (fonts, masthead eyebrow text, footer text) have working defaults you can leave alone. The skill ships with a --check diagnostic mode that previews the configuration and verifies the install before you render anything.
The more specific you make these configurations, the less generic the output will be.
Status
Early, working draft. Built and tuned for a single congregation in the evangelical Protestant tradition. Pending evaluation across other congregations and feedback from actual small group leaders. The pipeline adapts to adjacent traditions (Reformed Baptist, Anglican, Methodist, non-denominational evangelical) with the right configuration, but it assumes named-pastor expositional preaching and the specific category of “small group guide” as a thing your church does. If your tradition is liturgically very different, the underlying skill instructions may need to be edited rather than just configured.
Issues, forks, and pull requests welcome.