multica-ai/multica is the open-source managed-agents platform that sits behind the karpathy-skills phenomenon and the broader skill-bundle ecosystem now trending on GitHub. Multica positions coding agents as 'real teammates' — assign tasks, track progress, compound skills across the team — rather than as one-shot pair-programmers. It runs on top of any major model provider (Claude, GPT, Gemini, local models) and integrates with the agentskills.io SKILL.md standard so skill bundles trending on GitHub (Anthropic-Cybersecurity-Skills, ECC, andrej-karpathy-skills) can be installed and routed across team members. The repo has 33K stars overall and trended at +732 stars in the 24 hours of May 25 2026, alongside the broader Skills-paradigm convergence anchored by Pope Leo XIV's AI encyclical and Microsoft's agent-governance-toolkit on the same trending board.
Multica reframes coding agents from single-task pair-programmers to a managed team. The platform handles task routing, progress tracking, and skill propagation — when one agent learns a new SKILL.md pack, the platform makes it available to the rest of the fleet rather than locking that knowledge to a single chat session.
The platform’s core abstraction is the managed agent — a worker with a stable identity, a task queue, an audit log, and a shared skill library. Skills follow the agentskills.io SKILL.md standard, which means a skill written for Claude Code can run on a Multica agent backed by Codex, and a community skill bundle pulled from GitHub trending (like Anthropic-Cybersecurity-Skills or ECC) can be installed across the entire team with one command.
Engineering teams use Multica as a coordination layer for parallel agent work — multiple feature branches, multiple PRs, multiple bug investigations running concurrently with deterministic task tracking. Security teams use it as the orchestration substrate for the cybersecurity-skills bundle. Research teams use it for vertical-skill experiments (e.g., the scientific-agent-skills and academic-research-skills bundles AgentConn covered in the skills-go-vertical cycle).
GitHub Trending board on May 25 2026 is dominated by the Skills paradigm — multica/andrej-karpathy-skills (154K stars), affaan-m/ECC (192K stars), and Multica itself (+732 today). The convergence is not a coincidence: Skills as an interoperable spec has hit critical mass, and a managed-agents platform is the natural place to run a skill catalog of that scale.
Provider-agnostic. Works with Anthropic Claude Code, OpenAI Codex, Cursor agent mode, GitHub Copilot agents, and any self-hosted model with a chat-completions endpoint. Compatible with the SKILL.md standard, so any skill bundle on the skills directory race trending lane will install in Multica without rewriting.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.
Open-source AI pair programming tool that works in your terminal to edit code across your entire repository.
AWS's AI-powered coding assistant that helps developers build, deploy, and optimize applications on AWS with code generation and transformation.