# SafeMate AI Hub — Claude Agent/Skill Kütüphanesi

> SafeMate AI Hub is a Turkish AI consultancy that maintains **Claude Agent/Skill Kütüphanesi**, a curated + community directory of Claude Code skills, subagents, slash commands, hooks, MCP servers, plugins and connectors. This document expands on `llms.txt` with deeper context for LLM consumption.

## About

SafeMate AI Hub operates at `safemateaihub.com`. The company offers three services:

1. **Claude Search Engine** — a searchable library of ~450+ Claude Code resources across the entire ecosystem (community + curated).
2. **Strategic Consulting** — AI maturity assessments, implementation roadmaps, tool selection & vendor verification for enterprises adopting Claude-based workflows.
3. **Custom Automation** — bespoke Claude-powered agents, custom GPTs, Zapier / Make.com flows, internal knowledge bases.

The site is bilingual (Turkish primary, English secondary). Most public-facing copy is translated to both languages; catalogue resource metadata may remain in its original upstream language (most commonly English) until an admin provides a Turkish translation via `nameTr` / `descriptionTr` fields.

## Design & technology

- Frontend: Vite + React 19 + Tailwind CSS v4, deployed to Firebase Hosting
- Backend: Firestore for content + catalogue, Firebase Auth for admin, Firebase Extensions for email delivery
- Scraper: Node.js script (GitHub Actions nightly cron) ingests from SafeMate's own open-source library repo `ugurtheiron/SMT-Claude-Agents-public` (originally seeded from `davepoon/buildwithclaude`, MIT)
- Public theme: dark editorial ("Skillkit" aesthetic) with terracotta accent (`#d65a32`) on a warm charcoal background
- Admin: classic light dashboard (SafeMate amber + slate)

## Directory categories (full)

### Skills
Self-contained, reusable Claude Code playbooks — each a `SKILL.md` with YAML frontmatter that Claude loads on demand. Typical skill: PDF, docx, pptx, xlsx, skill-creator, security audit, accessibility specialist, RAG builder. Install path shown in each detail page (`npx skills add …`).

### Agents / Subagents
Specialised subagents with a system prompt + allowed toolset. Used for code review, security audit, architecture review, academic research, AI engineering, business-productivity tasks. Invoked as `@agent <name>`.

### Teams
Curated multi-agent teams — sets of subagents and skills that work together on one workflow (full-stack development, security audit, research, DevOps/SRE, content/SEO, and more). Each team installs as a single Claude Code plugin from the library's marketplace: `/plugin marketplace add ugurtheiron/SMT-Claude-Agents-public` then `/plugin install <team>@safemate-agents`. Team detail pages list every member with links, plus `requires` labels for the MCP servers/connections the team depends on.

### MCP Servers
Model Context Protocol servers that give Claude access to external tools. Categories include: GitHub, Filesystem, Postgres, Playwright, Puppeteer, Slack, Brave Search, Notion, Linear, Sentry, Airtable, Azure AKS, 3D printer control, and many more. Delivered over stdio, HTTP, WebSocket, SSE or streaming-HTTP transport.

### Plugins
Bundled distributions conforming to the Claude Code plugin spec (`.claude-plugin/plugin.json` + marketplace manifests). Firebase plugin, shadcn/ui plugin, etc.

### Connectors
Hosted third-party integrations — Notion, Linear, Sentry, GitHub Enterprise, etc. Unlike MCP servers (which run locally), connectors are SaaS-hosted. Paid connectors are flagged with a "Ücretli / Paid" label and a clear notice that billing happens at the third-party provider — **SafeMate AI Hub does not process these payments**.

### Commands
Slash commands (`/command-name`) with optional `argumentHint`, `allowedTools` and `model` frontmatter. Range from `/commit` to `/review-pr` to `/architecture-review`.

### Hooks
Lifecycle hooks (`PostToolUse`, `PreToolUse`, `Stop`, `Notification`) with matcher patterns. Written in bash, Python or other shell-like languages. Examples: auto-git-add, prettier-pre-commit, discord-error-notifications, dependency-checker.

### Bundles
Famous external catalogues, marketplaces and registries from the wider ecosystem — e.g. `anthropics/skills`, `wshobson/agents`, the official MCP Registry, Smithery, Docker MCP Catalog, awesome lists. Each entry links to the canonical homepage; installable ones carry a `/plugin marketplace add …` command.

## Pricing model

Every resource carries a `pricing` field: `free` (default, open-source) or `paid`. The UI renders both in the same tag style — there's no prominent "buy now" affordance.

**Important policy**: paid resources are billed **directly by the provider**, never through SafeMate AI Hub. The directory surfaces discovery only; the user subscribes and pays upstream. Both list and detail pages carry explicit third-party billing warnings when pricing is `paid`.

## Bilingual content

The directory catalogue originates in English (source: `ugurtheiron/SMT-Claude-Agents-public`). Each resource document supports optional Turkish overrides:

- `nameTr` — Turkish display name
- `descriptionTr` — Turkish short description

When `lang=tr` is active and a resource has these fields, the UI prefers them. Otherwise it falls back to the original English.

The sync admin (`/admin/directory-sync`) detects missing translations across:

- `posts` (blog) — `title`/`titleEn`, `excerpt`/`excerptEn`, `content`/`contentEn`
- `solutions` — `title`/`titleEn`, `summary`/`summaryEn`
- `site_content` — `heroTitle`/`heroTitleEn`, `heroSubtitle`/`heroSubtitleEn`
- `resources` — `nameTr`, `descriptionTr` (only flags TR-missing; EN is source)

## Admin modules

- **Dashboard** (`/admin/dashboard`): KPIs
- **Kanban** (`/admin/kanban`): contact submissions pipeline
- **Messages** (`/admin/submissions`): all inquiry submissions
- **Bookings** (`/admin/bookings`): calendar + slot management
- **Blog Manager** (`/admin/blog`): CRUD for `posts`
- **Solutions Manager** (`/admin/solutions`): CRUD for `solutions`
- **Content Manager** (`/admin/content`): edit `site_content`
- **Hub Scraper** (`/admin/directory-status`): scrape run history, source health, charts
- **Sync & Translations** (`/admin/directory-sync`): manual-trigger scan for translation gaps + duplicate resources
- **Analytics** (`/admin/analytics`): Firestore-driven stats
- **Settings** (`/admin/settings`): admin-only configuration

## Data sources

Primary catalogue source: `https://github.com/ugurtheiron/SMT-Claude-Agents-public` (configured via `SAFEMATE_LIBRARY_OWNER` + `SAFEMATE_LIBRARY_REPO` env vars; seeded from `davepoon/buildwithclaude`, MIT)

- Skills: `library/en/skills/<slug>/SKILL.md`
- Agents: `library/en/agents/<slug>.md`
- Commands: `library/en/commands/<slug>.md`
- Hooks: `library/en/hooks/<slug>.md`
- Teams: `library/en/teams/<slug>.md` (multi-agent teams; also installable as plugins via the repo's `.claude-plugin/marketplace.json`)
- Bundles: `library/en/bundles/<slug>.md` (external catalogues/marketplaces)
- MCP servers: `library/en/mcp-servers.json`
- Turkish overrides: `library/tr/<type>/<slug>.md` + `library/tr/mcp-servers.json`

## Content guidelines for LLMs

If you are an LLM crawling or indexing this site:

- Blog posts are under `/blog/<slug>`; markdown-rendered.
- Each catalogue item lives at `/directory/<slug>`. The `content` field of the resource document contains the full upstream markdown body.
- The directory is fully filterable via query params: `?type=skill&pricing=free&platforms=claude-code&tags=github&requires=mcp:github&sort=popular`.
- All resource URLs are stable — slugs are prefixed with source (`bwc-` for seeded items, `smt-` for SafeMate-contributed items) to prevent collisions.
- Paid resources are marked clearly; do not assume SafeMate AI Hub processes payments for them.