English | Русский | Español | Deutsch | Français | 中文 | 日本語 | Português | 한국어
A thin overlay on top of Superpowers (brainstorm, subagents, TDD, debug) and Get Shit Done (Spec → Plan → Execute) that closes the gaps those plugins leave for solo product builders.
For: solo founders and one-person engineering teams shipping real products with Claude Code.
Supported stacks: Laravel · Rails · Next.js · Node.js · Python · Go.
| Gap | What the toolkit adds |
|---|---|
| Multi-AI plan validation | /council — sends your plan to Gemini and ChatGPT in parallel for independent review. Works through CLI (gemini, codex) or direct API keys. Persona overlays, content-hash cache, cost gate, ru locale. |
| Product validation gate | product-thinking skill + /product-review — RIGID gate before code: target user, JTBD, pain intensity, success metric, distribution channel, structural advantage, unit economics with SaaS-graveyard floor, cheapest experiment with decision rule, top risk. Domain-configurable via ~/.claude/product-config.json. |
| Framework context | 7 ready-made CLAUDE.md templates (base + 6 stacks), auto-detected via artisan / next.config / go.mod / pyproject.toml / package.json. |
| Production safety net | cc-safety-net blocks destructive commands (rm -rf /, git reset --hard, etc.) at PreToolUse — even through obfuscation. Wired into the installer. |
| Token-cost control | RTK rewrites verbose dev-command output (git status, test runners) — 60-90% token savings. Combined hook with cc-safety-net. |
| Cost routing | better-model routes simple tasks to cheaper models. Auto-installed and integrated into the install lifecycle. |
| Symbol-aware code search | Serena (LSP, MIT, local) + ripgrep + claude-context (semantic vector). Default Layer-3 search stack. |
| Multi-CLI bridges | Auto-sync CLAUDE.md to GEMINI.md (Gemini CLI) and AGENTS.md (OpenAI Codex). Drift-detection at every install. |
| Integrations catalog | TUI installer for 31 MCP servers + 10 companion CLIs across 10 categories (Backend / Payments / Workspace / Project Management / …). Per-row scope. |
| Limit visibility (Pro/Max) | Statusline shows session/weekly usage — you can see when you're about to hit the wall. |
| Dependency dashboard (v6.2) | /update-deps — interactive TUI listing every tracked dependency (Layer 1/2/3) with installed-vs-latest. You pick what to update. |
| Post-install setup guide (v6.3) | Generates a local HTML page (/.claude/setup-guide.html) with per-MCP API-key walkthroughs and per-component config — only sections for what you actually installed. |
| Vendor functional changelog (v6.3) | /vendor-changelog — pins external vendors (Superpowers, GSD, Serena, RTK, …) at every release; diffs HEAD vs pin; classifies changes BREAKING/ADOPT/IGNORE/DEPRECATE. Auto-pin on release via .github/workflows/auto-pin-vendors-on-release.yml. |
| Brand-inspired UI tokens (v6.49) | /design-md — drop one of 71 brand-inspired DESIGN.md files (Vercel, Stripe, Airbnb, Apple, Tesla, …) into the project root so coding agents generate UI that matches a chosen visual language. Mirror pinned from voltagent/awesome-design-md. |
| Marketing skills (v6.51) | 40 marketing skills mirrored from canonical coreyhaines31/marketingskills: CRO, copywriting, SEO, pricing, paywalls, cold-email, churn-prevention, customer-research, programmatic-seo. Active skills_pins count 61. |
The headline value is curation. Everything is opt-in via TUI checkboxes — nothing is forced.
One command. Run it in a regular terminal inside your project folder (not inside Claude Code):
bash <(curl -sSL https://raw.githubusercontent.com/sergei-aronsen/claude-code-toolkit/main/scripts/install.sh)The installer shows a TUI checklist (Toolkit, Security, RTK, Statusline, Council, Bridges, Integrations) and detects whether superpowers and get-shit-done are already installed. If they are, it skips the files those plugins already provide and only ships the ~47 unique toolkit contributions.
For Claude Desktop users — install via marketplace:
/plugin marketplace add sergei-aronsen/claude-code-toolkit
Full step-by-step guide (with the install TUI walkthrough): docs/howto/en.md.
Use these from inside Claude Code after /plugin marketplace add:
| Command | What it does |
|---|---|
/update-toolkit |
Pull fresh toolkit content into .claude/ while preserving local edits. |
/update-deps |
Open the dependency dashboard (Layer 1/2/3 + MCP). Pick what to update. |
/council |
Send a plan to Gemini + ChatGPT for independent review. |
/product-review |
4-persona business review (skeptic + marketer + CFO + user-empath) — runs before technical planning. |
/vendor-changelog |
Diff pinned external vendors against HEAD; classify changes BREAKING/ADOPT/IGNORE/DEPRECATE. |
/learn |
Save the current decision as a scoped rule for future sessions. |
/audit |
Run one of the 7 framework-aware audits (security, performance, etc.). |
/debug |
4-phase systematic debugger: root-cause → pattern → hypothesis → fix. |
/setup-guide |
Regenerate the local HTML setup walkthrough for installed MCPs/components. |
/design-md |
Pick a brand-inspired DESIGN.md (71 brands) and drop it at the project root for UI agents. |
Full command list: docs/features.md.
The toolkit is a thin overlay organized in three layers:
- Layer 1 — toolkit content (templates, slash commands, components, skills, agents)
- Layer 2 — free base plugins (Superpowers, Get Shit Done, ru-text)
- Layer 3 — optional external tools (cc-safety-net, RTK, Serena, claude-context, better-model)
Full diagram: docs/architecture.md. For solo founders / non-developer product builders: docs/non-programmer-mode.md.
The --integrations flag (or /integrations after the first install) opens a TUI checklist with 31 servers across 10 categories. Pick only what your project needs — the rest stays untouched.
| Category | Servers |
|---|---|
| docs-research | context7 · firecrawl · notebooklm |
| backend | aws-cloudwatch-logs · aws-cost-explorer · cloudflare · dbhub · supabase |
| payments | stripe |
resend · mailgun |
|
| workspace | calendly · notion |
| project-management | jira · linear · youtrack |
| communication | slack · telegram |
| design | figma |
| dev-tools | magic · openrouter · serena · claude-context · playwright |
| monitoring | sentry · datadog · posthog |
Each server installs with per-row scope choice ([U] user / [P] project / [L] local). Project-scope writes credentials to <project>/.env (mode 0600) with auto-.gitignore; .mcp.json carries only ${VAR} substitution form. More: docs/INTEGRATIONS.md.
MIT — see LICENSE.