Event recap generator

1. The business decision

After every case competition, the club president sends a standard debrief — same sections, same tone, same length, every time. The problem: the person who takes notes is never the person who writes the recap, and turning messy bullet points into the finished document takes two to three hours. This workflow cuts that to one prompt. There's no code involved — the folder structure and CLAUDE.md do all the work.

2. Data available

• inputs/competition-invite-fall-2025.docx — official invitation from the organizers. Download
• inputs/notes.txt — raw notes from whoever attended. Download
• references/competition-recap-fall-2024.docx — last semester's approved recap, from a different competition. The format to match. Download

3. The build

Two rounds, about 20 minutes total.

Round 1 · Write the CLAUDE.md together

Don't write the CLAUDE.md alone. Instead, ask the AI to interview you:

I'm setting up a folder for generating recurring event recap
documents. I have a reference example and raw notes as inputs.

Before we write anything, ask me the questions you'd need
answered to write a good CLAUDE.md — the file that tells you
the rules for this project every time you open it.

Ask one question at a time.

The AI will surface things you hadn't thought to specify: whether certain sections are mandatory even if nothing happened in that area, how to handle names, whether to flag anything for human review before sending. Once you've answered, ask it to draft the CLAUDE.md and review it before saving. Then ask it to summarise what it sees in references/ — a quick sanity check before anything gets generated.

A well-written CLAUDE.md for this project looks something like:

# Event Recap Project

## Purpose
Generate a polished event recap document from raw notes,
matching the format in references/.

## Mandatory sections
The following sections must always appear, even if the notes
contain little or nothing for them:
- Event overview (date, location, format, team)
- Results
- Key learnings (minimum two)
- What we'd do differently
- Thank-yous (always include sponsors by name)

## Format rules
- Match the heading structure and approximate length of the
  reference document exactly
- Do not invent facts. If the notes are silent on something,
  write "[to confirm]" rather than guessing
- Names: use first name only for students, full name + affiliation
  for external contacts

## Output
Save the finished document to outputs/ as
  recap-[event-name]-[YYYY-MM].docx

Round 2 · Generate and review

With the CLAUDE.md in place and the format confirmed:

Now generate the recap using both files in inputs/ and the
reference document. Save the output to outputs/.

Read the output and check two things: every mandatory section is present, and any [to confirm] flags are things you can actually fill in. If a mandatory section is missing, the most common cause is a CLAUDE.md rule that was too vague — tighten it and re-run.

4. What good looks like

The most instructive thing you can do is run this workflow twice: once with a minimal CLAUDE.md and once with the full version above, using the same notes. The difference makes the case for why the instructions file matters more than the prompt.

A good output lands in outputs/, matches the reference on structure and length, and contains no invented facts — only [to confirm] placeholders where the notes ran dry.

5. Extensions

Once the base workflow is running, try these to go further:

• Run the workflow for three events in a row with different notes files. The folder stays the same — only inputs/ changes. This is the core payoff of the folder structure: the setup cost is paid once.
• Add a rule to CLAUDE.md asking the AI to flag any "lessons learned" items that appear across multiple events, once you have three or more recaps in references/. The folder becomes institutional memory.

The cross-event pattern analysis is a good lead-in to Module 6.

6. Skills used