gstack OpenClaw Office Hours Skill Review: YC-Style Questions in 16,964 Bytes
The native OpenClaw version of gstack's office-hours skill is a single 16,964-byte markdown file. That matters because the full Claude Code office-hours/SKILL.md in the same repo is 71,511 bytes, and its directory totals 185,440 bytes. The OpenClaw port keeps the product-interview core and drops most host-specific machinery.
TL;DR / Key Takeaways
gstack-openclaw-office-hoursis a 16,964-byte SKILL.md, about 4,241 estimated context tokens using bytes ÷ 4.garrytan/gstackhad 130,481 stars, 19,620 forks, 785 watchers, and 826 open issues on August 31, 2026.- The repo is MIT licensed, and GitHub code search found 119 files named
SKILL.mdin it during this run. - The inspected OpenClaw skill directory contains one file and no scripts, network calls, telemetry calls, credential reads, or shell snippets.
- Public discussion is mostly GitHub issue traffic; Hacker News had a gstack story with zero comments, so there is no broad HN verdict yet.
What the skill actually says
This skill tells an OpenClaw agent to behave like a YC office-hours partner. It does not ask the agent to code. The boundary is explicit:
"HARD GATE: Do NOT invoke any implementation, write any code, scaffold any project, or take any implementation action."
The best part is the product taste. Two short lines carry most of it:
"Specificity is the only currency. Vague answers get pushed."
"Interest is not demand. Waitlists, signups, 'that's interesting' ... none of it counts."
The workflow asks what the user is trying to do, maps the answer to startup mode or builder mode, then pushes on the named user, current workaround, demand proof, narrowest wedge, and concrete assignment. Startup mode is intentionally uncomfortable. Builder mode is softer, aimed at side projects, hackathons, learning, open source, and research. Both paths end in a design document, not implementation.
What it costs you in context
We did not run this skill through KissAPI or publish traffic-derived token counts. The estimate below is just file-size arithmetic: bytes ÷ 4 ≈ tokens.
| File | Bytes | Estimated context tokens | What it contains |
|---|---|---|---|
openclaw/skills/gstack-openclaw-office-hours/SKILL.md | 16,964 | ≈ 4,241 | OpenClaw-native office-hours workflow, startup diagnostic, builder brainstorm, design-doc output rules |
| Loaded set inspected | 16,964 | ≈ 4,241 | One markdown file; no pulled-in local references in this skill directory |
About 4.2k estimated tokens is reasonable for a planning stance that can redirect a project before code exists. It is also small compared with the full Claude Code office-hours skill. If you want product questioning inside OpenClaw chat, this is the lighter route.
What the community reports
The useful reception data is in GitHub issues, not listicles. The repo has active users filing detailed reports and maintainers responding. I also checked HN Algolia: the April 3 gstack submission exists, but had zero comments when checked.
| Source | Date | What was reported | Link |
|---|---|---|---|
| GitHub issue #2719 | 2026-08-28 | A Conductor user reported that full /office-hours never asked the six forcing questions across 22 invocations because open-ended questions did not fit the prose fallback. | issue #2719 |
| GitHub issue #542 | 2026-03-27; comment 2026-05-23 | Users reported that office-hours design docs can drift from Claude Code plan files, making downstream reviews read stale source-of-truth files. | issue #542 |
| GitHub issue #340 | 2026-03-22; comment 2026-07-15 | A user reported wireframe handoff problems: temporary files, headless display trouble, and downstream design-review skills not reliably using the generated image. | issue #340 |
| GitHub issue #2657 | 2026-08-20; reply 2026-08-22 | A long-time user showed office-hours tenure undercounting. Garry Tan verified the arithmetic and asked for the reconcile PR. | issue #2657 |
| GitHub issue #619 | 2026-03-29; comments through 2026-08-01 | Several users asked for provider-neutral "outside voice" support; one office-hours user said Codex and Gemini disagreed in a way that improved a product decision. | issue #619 |
| Hacker News Algolia | 2026-04-03 | The gstack repo was submitted as "Claude Code skills that turn one engineer into team of 20," but the story had zero comments when checked. | HN search |
Most complaints target the full Claude Code workflow around office-hours: AskUserQuestion fallback, plan-file authority, generated wireframes, and persisted profile state. The OpenClaw-native skill avoids some of that because it is just markdown, but the philosophy and downstream expectations are the same.
How to install it
The README lists native OpenClaw installation through ClawHub:
clawhub install gstack-openclaw-office-hours gstack-openclaw-ceo-review gstack-openclaw-investigate gstack-openclaw-retroFor the full Claude Code gstack setup, the README gives:
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack
cd ~/.claude/skills/gstack
./setupUse the first path if you want conversational OpenClaw methods. Use the second if you want Claude Code slash commands such as /office-hours, /review, /qa, and /ship.
Where it earns its keep, and where it doesn't
Use this when the next bad move would be building too soon. The skill is good at forcing a named user, a real pain point, a status quo competitor, and one concrete assignment. That is valuable because agents love turning vague ideas into polished plans.
My main criticism is the closing pitch. The skill ends with versions of a Garry note that point toward applying to Y Combinator. Sometimes that fits. Often it will feel like brand gravity inside the user's planning session. The issue tracker has already seen complaints around repeated YC suggestions in shared voice instructions, so this is not just my taste.
The second caveat is temperament. "Specificity is the only currency" is right for founder fog. It can be overkill for a toy project, a learning exercise, or an internal tool where the audience is already obvious. I would keep this skill on hand for product discovery. I would not make it the default for every feature plan.
FAQ
What does gstack-openclaw-office-hours do?
It gives an OpenClaw agent a YC-style office-hours workflow: classify the project, ask demand questions, challenge the premise, generate alternatives, and write a design document without coding.
How much context does it add?
The inspected SKILL.md is 16,964 bytes, about 4,241 estimated context tokens using bytes ÷ 4. This is not a traffic-derived session measurement.
Is it safe to install?
The inspected directory contains one markdown file and no scripts, shell snippets, network calls, telemetry calls, or credential access. Still read it first; a skill is instruction text handed to an agent.
Should I use this instead of full gstack?
For product interrogation inside OpenClaw chat, start here. For the full Claude Code sprint flow—office-hours into review, QA, ship, and release—install the full gstack repo and expect more moving parts.
Skill by Garry Tan and gstack contributors, MIT licensed. Source: github.com/garrytan/gstack. Quoted lines are short excerpts for review purposes. Review date: August 31, 2026. Community reception shifts over time; check the issue tracker before making install decisions.