Community Contributors

A 23-agent engineering and product team that ships from one session — two commands, zero meetings

tonone turns Claude Code into a full delivery team: 23 specialist agents (architect, reviewers, product, QA and the rest of the org chart) coordinated through 125 skills, so one session runs discovery, build, review, and ship without you playing project manager between tools. The mirror featured here is the curated cut — every one of its 114 agent definitions verifies clean at the kernel-strict frontmatter floor in the latest inventory run, preserved by the marketplace's never-clobber freeze. MIT-licensed and actively developed upstream; a frontmatter sweep for the skill tier is offered upstream and the roster works as-is today.

One session. Two commands. Full team. Zero meetings.

Persistent memory for Claude Code that actually gets written — measured, not promised

The mnemos skill closes the loop most memory tools leave open: getting anything recorded at all. A SessionStart hook prewarms a token-budgeted context block and a UserPromptSubmit hook pattern-matches correction-shaped prompts into a non-skippable capture directive — upstream measured capture jumping from a 7% baseline to 53%, with a +40% behavior lift on paired Claude runs. Corrections persist as structured tried/wrong_because/fix records, and three that cluster auto-promote into a skill deterministically, no LLM in the loop. The engine is a dependency-free single Go binary (~15 MB, pure-Go SQLite) keeping everything local. MIT-licensed, with all 8/8 required frontmatter fields verified at upstream HEAD.

A memory layer for AI coding agents — because Claude Code keeps forgetting, and that gets old fast.

Find the ~$27K/month leaking out of a $100K Databricks workspace — from the billing table, not a guess

The databricks-cost-leak-hunter skill audits a workspace for real-dollar cost leaks and emits a CFO-grokkable, dollar-ranked FinOps report. Every confirmed figure is computed from the customer's own system.billing.usage joined to list_prices — never an estimate. Four named leaks — idle clusters that never auto-terminate, scheduled jobs on All-Purpose compute (2–3× overpay), overprovisioned clusters, and the Photon premium paid without the speedup — each tagged confirmed / estimated / at-risk so a CFO never confuses billed waste with modeled savings. A deterministic Python ranker does the arithmetic (the LLM never eyeballs a number), and the databricks-workspace-mcp control plane turns each leak into a single-config-change fix. One of 24 skills in the Databricks pack.

A $100K/month Databricks workspace is likely burning ~$27,000/month — and every line is one config change.

Real mobile devices on demand — no emulators, no flaky CI

Kobiton's automate plugin gives Claude Code, Cursor, Codex, and Gemini CLI direct access to the Kobiton real-device cloud via remote MCP. 12 tools across three surfaces — Devices (list / reserve / status / terminate), Sessions (list / get / artifacts / terminate), and Apps (upload / confirm / list / get) — plus 3 specialist agents for device picking, Appium capability reconciliation, and session triage. One install, real iOS + Android hardware, no emulator drift.

Mobile testing that runs where the bugs actually live — on the device, not the simulator.

AI agents that drive your browser. 21,000 stars of Selenium-killer.

Skyvern automates browser-based workflows with vision-language models — point it at a page, describe the goal, and the agent navigates, fills forms, handles 2FA, and extracts data. Where Selenium and Playwright break the moment a page redesigns, Skyvern adapts because it sees the page like a human would. The CLI skill exposes Skyvern's task runner inside Claude Code: navigate, fill forms, extract data, handle logins, and stream back structured results. Pulled into the marketplace via daily external sync from Skyvern-AI/skyvern (AGPL-3.0). Sourced into the catalog by community contributor @mark1ian.

Automate browser-based workflows using LLMs and computer vision.

11-dimensional codebase cleanup with confidence scoring and build verification gates.

Born from a real cleanup session — 25,000 lines removed, 4 bugs caught — packaged into a reusable plugin. 11 specialized agents scan for dead code, AI slop, weak types, security vulnerabilities, N+1 queries, floating promises, circular deps, and more. Every finding gets a confidence score (HIGH/MEDIUM/LOW). LOW-risk dimensions auto-apply after build verification. HIGH-risk dimensions flag only — never touch your code without proof. One command, full codebase hygiene.

8 parallel agents, 25K lines removed, 4 bugs found. Now it's a plugin anyone can run.

Your dashboards don't talk. This skill does.

A push-based analytics intelligence team — 9 specialist agents that fetch your data, detect anomalies, analyze funnels, verify claims, and deliver narrative reports. Powered by self-hosted Umami via MCP — agents call the Umami API in real time across your entire site portfolio. GA4 as fallback. Three tiers: /analytics for a 30-second pulse, /analytics medium for a daily brief, /analytics full for a deep cross-site dive with adversarial verification.

I stopped checking dashboards. Now my analytics team wakes me up when something matters.

Ghost tokens eat your context. Find them. Fix them. Prevent rot.

My Claude Code kept leaking tokens. So I built this. Now my limits last longer, context stays fresh, and I have full visibility and control over my usage & setup.

The AI skills that fully replaced a $4k/mo executive assistant

Research meeting attendees, gather email history, create meeting briefs, and manage action items via Todoist — the complete EA workflow automated by Auth0's former VP of Marketing/Growth.

These skills fully replaced my human executive assistant.

The meta-tool that builds and grades other skills

30 skills that make Cursor actually intelligent

Professional crypto portfolio tracking with real-time PnL