The Claude Code power-up you didn't know you needed (but absolutely do).
Goldy is a globally-installed toolkit that straps a jetpack onto Claude Code. It ships 31 skills, 19 agents, safety hooks, slash commands, and third-party integrations — all wired up through a slick terminal UI. No per-project config. No YAML nightmares. Just install and go.
- A planning engine (
/goldy) that writes Gold Standard project plans with stack-aware architecture, session memory, and resume capsules — so Claude stops winging it and starts engineering - A phase executor (
/goldy-loop) that runs those plans step-by-step with git worktrees, checkpoints, circuit breakers, and deep audits — like CI/CD but for your AI pair programmer - An interactive TUI installer built in Go + Bubbletea, because clicking through install wizards is so 2004
- A global toolkit — install once, works everywhere, follows you across every project like a loyal golden retriever
- 31 Skills — SEO (13 flavors), TanStack/React frontend, dev tools, brainstorming, pricing strategy, UI/UX design intelligence, and more
- 19 Agents — GSD workflow agents (plan, research, execute, verify, debug), SEO specialists, a markdown rewriter, and a pair programmer reviewer who actually reads your code
- 8 Hooks — Notifications, TTS (your terminal can talk now), safety guards against dangerous file deletions, credential leak prevention, and context monitoring
- 4 Extra Commands —
/agentfor orchestration,/plannotator-reviewfor interactive code review,/revise-claude-mdfor session reflection,/global-updateto refresh everything - 3rd Party Integrations — GSD (structured project management), Claude-Mem (persistent team memory), Plannotator (plan review UI)
- Gold Standard Plan Template — A battle-tested plan skeleton with 14 mandatory sections, because "just figure it out" isn't a spec
- You type
/goldy build me an auth systemin any Claude Code session - Goldy detects your stack (React? Next.js? Svelte? It knows 13 profiles), then generates a comprehensive plan with phases, requirements, test criteria, and acceptance targets
- You review the plan. You're still the boss here.
- You run
/goldy-loop --plan plans/auth-plan.mdand Goldy executes each phase in an isolated git worktree, runs lint + typecheck + tests before marking anything done, and hands you a completion report with actual evidence - If something breaks mid-flight, the circuit breaker kicks in (3-state: CLOSED → HALF_OPEN → OPEN) and Goldy stops before it makes things worse — unlike that one coworker
- Sessions are resumable. Close your laptop, go touch grass, come back, and Goldy picks up exactly where it left off
curl -fsSL https://raw.githubusercontent.com/SacredTexts/goldy/main/install.sh | bashnpm install -g goldy-cligit clone git@github.com:SacredTexts/goldy.git ~/.goldy
cd ~/.goldy && bash install.shPrivate repo? Use a GitHub token:
# With gh CLI (recommended)
gh auth token | GOLDY_TOKEN=$(cat) bash -c \
'curl -fsSL -H "Authorization: token $GOLDY_TOKEN" https://raw.githubusercontent.com/SacredTexts/goldy/main/install.sh | bash'
# With a personal access token
GOLDY_TOKEN=ghp_xxxx curl -fsSL -H "Authorization: token $GOLDY_TOKEN" \
https://raw.githubusercontent.com/SacredTexts/goldy/main/install.sh | bash- Interactive menu —
bash install.sh— browse and pick what you want - Install everything —
bash install.sh --all— the "yes to all" approach - Core only —
bash install.sh --core— just the skill, commands, and hooks - Cherry pick —
bash install.sh abc— install components a, b, and c by letter - Info mode —
bash install.sh --info— read descriptions before committing
| Location | What lives there |
|---|---|
~/.goldy/ |
Cloned repo (source of truth) |
~/.claude/skills/goldy/ |
Symlink to the repo |
~/.claude/commands/ |
/goldy and /goldy-loop slash commands |
~/.claude/agents/ |
19 agent definitions |
~/.claude/hooks/ |
Safety hooks and notifications |
~/.claude/GOLD-STANDARD-SAMPLE-PLAN.md |
The plan template |
| Flag | What it does |
|---|---|
--plan <file> |
Point to a specific plan file |
--max-iterations N |
Override the loop limit (default: 10, for the cautious) |
--commit-phase |
Auto-commit after each phase completes |
--diagnostics |
Enable diagnostic bundles for debugging |
--commands |
Print the full command reference |
cd ~/.goldy && git pull && bash install.shOr just run /global-update from inside Claude Code. It updates itself. The future is now.
make -C ~/.goldy uninstallNo hard feelings.
- Python 3.8+ (pure stdlib — zero pip dependencies, because dependency hell is real)
- Claude Code CLI (
npm install -g @anthropic-ai/claude-code) - Git (for worktree isolation in
/goldy-loop)
See CONTRIBUTING.md for the development workflow.
Private — SacredTexts team use only.