vcstack

Personal Claude Code skill stack. Forked from gstack by Garry Tan, customized for my workflow.

36 slash commands that turn Claude Code into a virtual team of specialists (CEO, eng manager, designer, reviewer, QA, release engineer), plus two binaries (browse for real-browser QA, design for GPT Image mockups).

Install

git clone https://github.com/vickcharles/vcstack.git ~/.claude/skills/vcstack
cd ~/.claude/skills/vcstack && ./setup

Restart Claude Code and type / to see the registered skills.

Requirements: Claude Code, Git, Bun v1.0+.

Typical workflow

/office-hours       → design doc from a raw idea
/autoplan           → CEO + Eng + Design + DX review in one command
<implementation>    → Claude writes the code
/review             → catch bugs before shipping
/qa <url>           → real-browser test, fix + regression tests
/ship               → tests, commit, push, PR

Skills (36)

Think and plan

Skill	Role	What it does
/office-hours	YC Office Hours	6 forcing questions that reframe the product before you write code
/plan-ceo-review	CEO	Find the "10-star product" hiding in the request. 4 scope modes
/plan-eng-review	Eng manager	Lock architecture, data flow, edge cases, tests
/plan-design-review	Senior designer	Scores each dimension 0-10, edits the plan until it's a 10
/plan-devex-review	DX lead	Explores dev personas, TTHW benchmarks, traces friction
/design-consultation	Design partner	Builds a design system from scratch, writes DESIGN.md
/autoplan	Review pipeline	Runs CEO + design + eng + DX automatically in one command

Design

Skill	What it does
/design-shotgun	Generates 4-6 AI mockups, opens a comparison board, iterates visually
/design-html	Mockup → production HTML/CSS with Pretext (computed layout, 30KB)
/design-review	Visual audit of a live site + atomic fixes with screenshots
/devex-review	Live onboarding test: navigates docs, times TTHW

Build and debug

Skill	What it does
/investigate	Root-cause debugging. Iron Law: no fix without investigation. Stops after 3 tries
/learn	Persistent cross-session memory. Review, search, prune learnings

Review

Skill	What it does
/review	Staff engineer. Bugs that pass CI but blow up in prod. Auto-fixes the obvious
/codex	Independent second opinion via OpenAI Codex CLI (review, challenge, consult)
/cso	Chief Security Officer. OWASP Top 10 + STRIDE. 8/10+ confidence gate

Test

Skill	What it does
/qa	Finds bugs, fixes them, generates regression tests, re-verifies
/qa-only	Same audit as /qa, report only, no code changes
/benchmark	Core Web Vitals, load times, bundle size. Before/after comparison
/health	Code quality dashboard (typecheck + lint + tests + dead code)

Ship

Skill	What it does
/ship	Sync main, tests, coverage audit, push, open PR. Bootstraps test framework if missing
/land-and-deploy	Merge PR, wait for CI + deploy, verify prod
/canary	Post-deploy monitoring: console errors, perf regressions
/document-release	Syncs all project docs with the shipped diff

Reflect

Skill	What it does
/retro	Weekly retro. /retro global scans across all projects and AI tools
/checkpoint	Save/restore work state across sessions or worktrees

Real browser

Skill	What it does
/browse	Persistent headless Chromium CLI. $B <cmd>, ~100ms per call
/open-vcstack-browser	Headed Chromium with sidebar extension and stealth
/connect-chrome	Connect to your real Chrome for direct control
/setup-browser-cookies	Import cookies from your real browser into the headless session
/pair-agent	Share the browser with another AI agent via tunnel + isolated tabs

Safety

Skill	What it does
/careful	Warnings before rm -rf, DROP TABLE, force-push
/freeze / /unfreeze	Lock edits to a single directory
/guard	/careful + /freeze combined

Setup

Skill	What it does
/setup-deploy	Detects deploy platform for /land-and-deploy

Which review do I use?

Building for...	Before code	After shipping
End users (UI, web, mobile)	/plan-design-review	/design-review
Developers (API, CLI, SDK)	/plan-devex-review	/devex-review
Architecture (data flow, perf)	/plan-eng-review	/review
All of the above	/autoplan (auto-detects which apply)	—

How it works under the hood

your slash command            → Claude Code reads <skill>/SKILL.md
SKILL.md says "run $B..."     → browse/dist/browse binary (Playwright/Chromium)
SKILL.md says "log it"        → bin/vcstack-* writes to ~/.vcstack/

• Skills = Markdown prompts generated from .tmpl files (source of truth)
• Binaries = browse/dist/browse, design/dist/design, bin/vcstack-global-discover
• Chrome extension = extension/ (sidebar for /open-vcstack-browser)
• Global state = ~/.vcstack/ (config, analytics, learnings, sessions)

Full stack details in ARCHITECTURE.md.

Development commands

bun install                # deps
bun test                   # validate skills, <2s
bun run dev <cmd>          # browse in dev mode
bun run gen:skill-docs     # regenerate SKILL.md from .tmpl
bun run build              # gen docs + compile binaries
bun run skill:check        # skill health dashboard
bun run dev:skill          # watch mode: auto-regen + validate

Iteration loop

1. Edit a <skill>/SKILL.md.tmpl.
2. Run bun run gen:skill-docs to regenerate the SKILL.md.
3. If ~/.claude/skills/vcstack is a symlink to this repo, changes are live in the next Claude Code session.

Check the symlink: ls -la ~/.claude/skills/vcstack.

Troubleshooting

• Skills don't show up: cd ~/.claude/skills/vcstack && ./setup
• /browse fails: cd ~/.claude/skills/vcstack && bun install && bun run build
• Stale install after rename: remove old skill folders from ~/.claude/skills/ and reinstall

Docs

Doc	Covers
ARCHITECTURE.md	Design decisions and internals
BROWSER.md	Full /browse command reference
CONTRIBUTING.md	Dev setup, testing
CHANGELOG.md	What changed in each version
docs/skills.md	Per-skill deep dives

License

MIT. See LICENSE.