# Office Hours > Run a YC-style office hours session before writing code — six forcing questions that surface wrong premises, missing constraints, and better approaches so you build the right thing the first time. - Skill: `manastalukdar/office-hours` (Agent Skill) - Install (CLI): `npx skillmds@latest add manastalukdar/office-hours` - Raw SKILL.md: https://api.skillmd.com/api/skills/manastalukdar/office-hours/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: manastalukdar (https://skillmd.com/u/manastalukdar) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/manastalukdar/office-hours --- # Office Hours Stop. Before writing a line of code, I'll put your plan through six forcing questions that surface wrong premises, hidden constraints, and better approaches. Inspired by gstack's office-hours skill and Garry Tan's YC office hours format. The goal is not to block progress — it is to spend 5 minutes now to avoid spending 5 hours on the wrong thing. ## Token Optimization **Expected range**: 400–1,000 tokens (interview + synthesis), 100 tokens (user has clear answers, no reframing needed) **Patterns used**: Progressive disclosure (one question at a time), early exit (user explicitly skips with `/office-hours --skip`) **Early exit**: If invoked with `--skip`, acknowledge and proceed without questions. ## When to Use Invoke before: - Starting a new feature or significant refactor - Writing a plan (`/write-plan`) - Any task that would take more than 2 hours of engineering work - When you feel uncertain about scope or approach ## The Six Forcing Questions Ask each question one at a time. Wait for the user's response before proceeding to the next. --- **Question 1 — What problem are you actually solving?** Describe the problem in one sentence without mentioning the solution. If you find yourself describing implementation details, you are not describing the problem yet. *Purpose: Separates problem definition from solution assumption.* --- **Question 2 — Who specifically has this problem, and how do you know?** Name a real user or situation. If the answer is "everyone" or "developers generally," the problem is not specific enough. *Purpose: Forces concrete user grounding vs. abstract feature building.* --- **Question 3 — What is the simplest thing that could possibly work?** Not the best solution — the simplest. What is the minimum version that would validate whether the problem is real? *Purpose: Surfaces whether the planned approach is over-engineered.* --- **Question 4 — What assumptions are you making that could be wrong?** List at least two. These are beliefs about the user, the system, or the environment that you have not verified. *Purpose: Makes hidden risk explicit before it becomes a bug.* --- **Question 5 — What would make this a bad idea?** Under what conditions is this the wrong approach entirely? What would you need to see to abandon the plan? *Purpose: Forces adversarial thinking while the cost of changing direction is low.* --- **Question 6 — What will you cut if this takes 3× longer than expected?** Scope always expands. Define what is non-negotiable vs. what is nice-to-have before you start. *Purpose: Prevents scope creep from killing the project.* --- ## Synthesis After all six answers, produce a one-page summary: ``` Office Hours Summary Problem: User: Minimum viable approach: Top risks: Kill conditions: Non-negotiable scope: Nice-to-have (cut first): Recommendation: ``` **Recommendations:** - **Proceed** — answers are clear, approach is well-grounded - **Reframe** — answers reveal the plan needs adjustment before starting; suggest a revised framing - **Stop** — the problem is not real, the solution is wrong, or the assumptions are too fragile; explain why ## Edge Cases - **User skips a question**: Record as `[skipped]` and move on; do not block - **User already has a plan**: Run questions anyway — the plan may change, and that is the point - **Small tasks (under 30 minutes)**: Still worth asking Questions 1 and 3; skip the rest - **Repeated invocation on the same task**: Show the prior summary and ask which answers have changed