Research Pulse — 2026-06-07-gstack

# Research Pulse — 2026-06-07 (gstack)
_Brown Biotech tooling track · manual curation by Demis_

## TL;DR
[gstack](https://github.com/garrytan/gstack) (Garry Tan / YC, **107,863⭐**, MIT) is a 23-skill Claude Code setup.
Two patterns are worth back-porting: **6 forcing questions** (from `/office-hours`) and the **10-star MVP
finder** (from `/plan-ceo-review`). Claude-Code-specific runtime; pattern is portable to Hermes.

### What gstack is
_Brown Biotech relevance: structured skill-set pattern; YC-grade intake/plan prompts; GBrain persistent memory_

- [gstack on GitHub](https://github.com/garrytan/gstack) — 23 opinionated skills, TypeScript, MIT
- [README — "Same person. Different era. The difference is the tooling."](https://github.com/garrytan/gstack#readme)
- [GBrain: persistent memory + 3 trust tiers (read-write / read-only / deny)](https://github.com/garrytan/gstack/blob/main/USING_GBRAIN_WITH_GSTACK.md)
- [SKILL.md.tmpl](https://github.com/garrytan/gstack/blob/main/SKILL.md.tmpl) — the skill-spec template
- All 23 skills: `office-hours`, `plan-ceo-review`, `plan-eng-review`, `plan-design-review`, `plan-devex-review`, `design-shotgun`, `design-html`, `qa`, `qa-only`, `ship`, `document-release`, `learn`, `retro`, `pair-agent`, `browse`, `scrape`, `context-save`, `context-restore`, `careful`, `guard`, `freeze`, `unfreeze`, `health`, `ios-*` (mobile subset)

### The 6 forcing questions (back-port candidate)
_Source: `office-hours/SKILL.md` Phase 2A — "YC Product Diagnostic" mode_

| # | Question | What to push for |
|---|---|---|
| **Q1** | Demand reality | Specific behavior — paying, expanding, panic-on-outage. Not waitlists. Not "VCs are excited." |
| **Q2** | Status quo | The cobbled-together workaround users are *already* living with. If "nothing," pain is too low. |
| **Q3** | Desperate specificity | A *name*, a *role*, a *consequence*. Not a category. "Email a category" is impossible. |
| **Q4** | Narrowest wedge | Smallest version someone pays for **this week**. One feature. One workflow. One email. |
| **Q5** | Observation & surprise | Have you *watched* a real user struggle? Demos are theater. Surveys lie. The gold is surprise. |
| **Q6** | Future-fit | In 3 years — does this become more essential or less? What shifts, and why does it help you? |

**Smart routing by product stage** (don't ask all six):
- Pre-product → Q1, Q2, Q3
- Has users → Q2, Q4, Q5
- Has paying customers → Q4, Q5, Q6
- Pure infra → Q2, Q4 only

**Anti-sycophancy rule:** never say "That's an interesting approach" or "There are many ways to think about this."
Take a position. State what evidence would change it.

### Plan-CEO-Review (back-port candidate)
_Source: `plan-ceo-review/SKILL.md` — finds the 10-star version + narrowest real MVP_

The pattern: given a fuzzy concept, output (a) the 10-star ceiling, (b) the smallest version that ships this week
with real money behind it, (c) what to *cut* to reach that MVP, (d) the decision the founder can act on today.
**Push back once, then push again.** The first answer is the polished one. The real answer is 2–3 pushes in.

### Brown Biotech relevance
_Brown Biotech relevance: existing 14 skills already cover 60% of gstack; the forcing-Q + plan-ceo-review are the missing 30%_

| gstack skill | Brown Biotech equivalent | Gap |
|---|---|---|
| `/office-hours` (6 Q) | `/intake` (triage + route) | Forcing-Q pattern missing |
| `/plan-ceo-review` (10-star MVP) | `/service-brief` (lane + CTA) | 10-star / narrowest-MVP pattern missing |
| `/plan-eng-review` | `/research-scan` (evidence summary) | — |
| `/document-release` | `/launch-check` | — |
| `/design-shotgun` (4-6 variants) | — | **New — design iteration workflow** |
| `/qa` (browser click) | `/dogfood` (partial) | Browser regression gap |
| `/ship` (release gate) | `/launch-check` | — |
| GBrain (persistent memory) | Hermes Agent memory layer | Different stack, similar idea |

### Decision: **back-port 2 patterns, skip 21 skills**

**Back-port (this turn):**
- 6 forcing Q → enrich `/intake` with a `--deep` flag that runs the diagnostic
- 10-star MVP finder → enrich `/service-brief` with a CEO-review subsection

**Skipped / deferred:**
- gstack's `bin/`, `claude/`, `codex/` runtime — Claude-Code-specific, not Hermes-portable
- GBrain file format — Hermes has its own memory layer; cross-pollinate via Notion instead
- `/design-shotgun` — design variant workflow; revisit when peptide-service or biostatx ships UI
- `/qa` browser regression — covered by `dogfood` skill in Brown Biotech stack
- Mobile `ios-*` skills — out of scope

### Verification
- `git ls-remote https://github.com/garrytan/gstack` → exists, 107,863 stars, MIT
- `SKILL.md` 48KB auto-generated from `SKILL.md.tmpl` (regen via `bun run gen:skill-docs`)
- `office-hours/SKILL.md` 88KB, Phase 2A confirmed at line ~53,225
- **Authorship caveat:** the viral tweet about gstack is from `@noisyb0y1` (a promoter), not Garry Tan. Repo is real;
  the 18-yr-old Cursor hackathon story is real but uses Cursor + Supabase + Vercel, not Hermes.

**Back-port actions executed (parallel):**
- `~/.claude/skills/intake/SKILL.md` — added `--deep` mode with 6 forcing Q + smart routing
- `~/.claude/skills/service-brief/SKILL.md` — added `## CEO-Review mode` (10-star + narrowest MVP)

**Pattern source files (read-only):**
- `/tmp/garrytan-gstack-readonly/office-hours/SKILL.md` (88KB)
- `/tmp/garrytan-gstack-readonly/plan-ceo-review/SKILL.md` (to read next)