Compose workflows, govern AI, and keep the human gate.

We build in the open. Plugins, skills, and MCP servers from the PromptMetrics team — the same tools behind our AI Fluency cohorts and Human Layer coaching — released for everyone, with the human review step designed in.

Every repo here shares one rule: a human approves every write before it fires. No autonomous action. The skills are plain markdown — read them, fork them, maintain them without us.

MCP server

• hubspot-mcp — 76 domain tools for HubSpot CRM. Every write produces a preview you approve or reject, with undo snapshots and an audit log.

Claude Code plugins

• hubspot-claude — HubSpot CRM admin. 44 sub-agents, 79 tools, every write gated behind a preview and an explicit approve.
• agent-patterns — matches your system against the live Encyclopedia of Agentic Coding Patterns (~304 entries) and returns the few that apply — leading with the gaps: the destructive path with no approval policy, the router with no eval gate.

Skills (plain markdown)

• skills — the open registry of Claude Agent Skills. Free to install and read. Start with qbr-prep, a read-only QBR drafter.
• coach-skill — coaches you through a real soft-skills conversation: feedback, pushback, a hard client or team talk. Built on The Human Layer method.
• habit-layer — one sharp question before Claude runs a real task. A tap on the shoulder, timed to the moment an old habit would take over.
• newsletter-skill — turns a newsletter brief into a Loops.so email, previews it, and sends only after two human approval gates. Never auto-fires.
• video-skill — plans Kurzgesagt-style explainer videos: script, storyboard, visual design spec. No render.

Something new lands every couple of weeks.

There's no demo. There's just the code.

A free skill that does most of a job isn't a lost sale. It's the proof — the proof that disarms a burned operator, and that wariness is well-earned, because the market is full of vendors who oversold. Don't take our word for it. Install one and watch it run.

What a markdown file can't do is the part we sell — connect it into your stack, govern who approves what, document the data flow, manage the change inside your team, and build the human fluency to keep it running.

More on the why, and what's coming next: promptmetrics.dev/open-source.

Two people. Ranya runs the human-skills side; Izzy runs the technical infrastructure. The human skills without the technical infrastructure is coaching. The infrastructure without the human skills is a tool rollout. PromptMetrics is what happens where they meet.

We're building our first case studies with our first customers right now. If you need someone else to have gone first, we're not the right fit yet — and we'll tell you that up front.

Tell us what your team needs and we'll consider building it in the open.

→ promptmetrics.dev · open-source · /promptmetrics/skills

_{Most repos are MIT. The Habit Layer is CC BY-NC-SA 4.0. Nothing here is a black box.}