yc-office-hours for Claude Code: a business mentor skill that asks before you build

yc-office-hours is a skill that tests a business decision with mandatory questions and saves the conversation as mentoring notes. It is a single skill with no public repository, so this post covers what it does, who should build one, the layout to start from, and the part you have to rewrite before it is useful to anyone but its author.

What problem it solves

Agents are eager to build. Describe a product idea to Claude Code and it will often start scaffolding before anyone has asked whether customers want the thing. yc-office-hours reverses that order. It checks the problem before the solution, prohibits code and scaffolding outright, and produces exactly one artifact: a set of notes with a single assignment for what to do next.

It also blocks flattery by rule. Remarks such as “That’s an interesting approach” are prohibited, and every answer has to state a position along with the evidence that would change it. The skill does not ask every question every time. It picks three of its six mandatory questions based on whether you have only an idea, some users, or paying customers, and it skips later questions that earlier answers already covered.

Who it fits

The chapter rates four situations. Three are enough to decide.

Situation	Fit	Reason
A solo founder evaluating a new product	Good fit	Probes demand and the initial wedge
Deciding whether to accept teaching or contract work	Good fit	Project acceptance mode and personal criteria
Needing code or documents immediately	Poor fit	Implementation is prohibited

Considering whether to build your own work automation sits in the middle. The skill handles it, but with collaborative questions rather than the interrogation it applies to new products.

Build it and run it

Because there is no public repository, you write the skill yourself. This is the structure the chapter gives.

# Example folder layout
~/.claude/skills/yc-office-hours/
  SKILL.md    # frontmatter(name, description, argument-hint, model) + body
# Body section order
#   Phase 0 Load context, Phase 1 Classify the agenda
#   Operating principles, anti-flattery rules
#   Phase 2A Mandatory questions, Phase 2B Builder mode
#   Phase 3 Challenge assumptions, Phase 3.5 Second opinion
#   Phase 4 Generate alternatives, Phase 5 Notes and assignment

It activates when you call it by name or ask for a review of a business decision.

Review [idea or decision to evaluate]

The run follows a fixed sequence. The skill first classifies the decision as a new idea, project acceptance, IP investment, or automation project, and identifies the product stage. It then asks its mandatory questions and pushes until the answers become specific and uncomfortable, moving on only after you agree to the shared assumptions. Finally it compares a minimum viable version against an ideal structure with equal weight, recommends one, and writes the notes for the option you choose.

What trips people up

The decision criteria belong to the author’s business. The skill body contains numbers such as a minimum effective hourly rate, a monthly cap on income-generating hours, and a maximum share of revenue from any single platform. If you copy the skill without replacing them, the mentor will judge your contract offers against someone else’s economics. Rewrite those thresholds first, and change the note storage path to a folder of your own while you are in the file.

Where this comes from

This post adapts part of chapter 01-05 of The Claude Code & Codex Skills Guidebook. The full chapter covers the six mandatory questions and the assumption-challenge lenses, such as working backward and classifying decisions by whether they are reversible, and the external files and tools the skill reads, including the codex exec second opinion in Phase 3.5, and the book has 52 more skills like this one. Read the full chapter in The Claude Code & Codex Skills Guidebook on Leanpub.