The gstack Directory
Every command in the toolbelt, what it does, and where the course teaches
it. Installed in Session 1 by pasting
https://github.com/garrytan/gstack into Claude Code. Don't read this top
to bottom, come back as each row becomes yours.
Every session, both directions
| Command |
What it does |
Learned |
/context-save |
Save where you got to before closing a session |
S1 |
/context-restore |
Pick up exactly where you left off |
S2 |
/gstack-upgrade |
Update the toolbelt (run weekly) |
S1 |
Safety (on from day one)
| Command |
What it does |
Learned |
/careful |
Warnings on destructive commands, the seatbelt |
S1 |
/guard |
Full safety mode: warnings + edits fenced to a folder |
S6 |
/freeze / /unfreeze |
Fence edits into one folder / formally move the fence |
S6 |
Deciding what to build
| Command |
What it does |
Learned |
/office-hours |
Pitch your idea to a hard-question investor, the business case |
S4 |
/spec |
Turn intent into a precise spec, interviews you, five phases |
S2 (slice), S4 (system) |
/plan-tune |
Calibrate how the review skills question you |
S5 |
/plan-ceo-review |
Gate: is it worth building? |
S5 |
/plan-eng-review |
Gate: is it technically sound? |
S5 |
/plan-design-review |
Gate: will a human enjoy it? |
S5 |
/plan-devex-review |
Gate: will it be pleasant to build and change? |
S5 |
/autoplan |
All four gates, auto-adjudicated, the board you later sit yourself in S5 |
S2 |
Design
| Command |
What it does |
Learned |
/design-consultation |
Research + a full design system, with previews |
S5 |
/design-shotgun |
Several design variants on a comparison board, pick one |
S3 |
/design-html |
The chosen design as production-quality HTML/CSS |
S3 |
/design-review |
Designer's-eye QA of the real thing, finds and fixes |
S8 |
/diagram |
English in, real diagram out (editable afterwards) |
S5 |
Building and debugging
| Command |
What it does |
Learned |
/investigate |
Scientific debugging: observations → root cause before any fix |
S6 |
/scrape |
Pull data out of a web page |
S3 |
/skillify |
Turn your last scrape into a permanent one-word command |
S3 |
/setup-gbrain / /sync-gbrain |
Long-term project memory, CLAUDE.md's index principle at codebase scale |
S7 |
Testing and quality
| Command |
What it does |
Learned |
/qa-only |
Robot tester, report only, a gate wants findings recorded |
S1, S8 |
/qa |
Robot tester that fixes as it finds, agile wants it done |
S2 |
/cso |
Chief Security Officer review of the whole project |
S8 |
/health |
Code-quality dashboard, the X-ray |
S8 |
/benchmark |
Performance baseline + regression detection |
S8 |
/devex-review |
Audit of your repo as a workplace |
S8 |
/codex |
Second opinion from a different AI model entirely |
S8 |
/browse |
The fast headless browser under /qa, /scrape, /benchmark |
S2 (seen) |
/connect-chrome |
A visible browser the agent can drive |
S1 |
/setup-browser-cookies |
Lend the agent's browser your logins (understand it before running it) |
S3 |
Shipping and watching
| Command |
What it does |
Learned |
/review |
Pre-landing review of a change, the review step inside /ship, standalone |
S8 |
/ship |
Tests → review → version → changelog → commit → push |
S1, S8 |
/setup-deploy |
One-time deployment configuration |
S2 |
/land-and-deploy |
Take a change through checks and out to the live site |
S2 |
/canary |
Watch the deployment after it ships, ship and observe |
S2 |
/landing-report |
Dashboard of everything shipped and queued |
S8 |
iPhone apps (the capstone)
| Command |
What it does |
Learned |
/ios-sync |
Install the debug bridge, scaffolding that lets the agent see into your phone |
S6 |
/ios-qa |
Robot tester driving the app on your real device |
S6 |
/ios-fix |
Autonomous iOS bug fixer (you re-run QA after, evidence, always) |
S8 |
/ios-design-review |
Visual design audit on the actual hardware |
S8 |
/ios-clean |
Strike the scaffolding: remove the debug bridge at closure |
S8 |
Documents and memory
| Command |
What it does |
Learned |
/document-generate |
Write a project's missing documentation |
S7 (homework) |
/document-release |
Update docs for what actually shipped |
S8 |
/make-pdf |
Any markdown into a polished, printable PDF |
S8 |
/retro |
Structured retrospective on how you worked |
S3, S8 |
/learn |
Bank lessons where future sessions will find them |
S3, S8 |
The one you only hear about
| Command |
What it does |
Status |
/pair-agent |
A remote agent paired with your browser, directing a team |
S8, the finale |
/benchmark-models |
How the gstack authors test their own toolbelt across AI models |
Awareness only, meta-tooling, not course material |
Every one of these commands is a markdown document (Session 3's reveal).
When you're curious how one works, read it.