looplia

Looplia Core

Skills-first agentic workflow CLI — compose AI skills and workflows for any task.

Looplia Core is an agentic workflow platform powered by the Claude Agent SDK. It provides a skills-first architecture where workflows declare which skill to execute and provide natural language missions, letting Claude determine the best approach.

Current focus: Content writing workflows (summarization, idea generation, writing kit construction)

Vision: A universal CLI for AI-powered workflows — one tool, many domains, powered by composable skills.

v0.7.4 Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│  CLI Commands                                                               │
│  └─ looplia init     → Initialize workspace + install marketplace skills    │
│  └─ looplia run      → Execute workflow on content                          │
│  └─ looplia build    → Build workflow from natural language                 │
│  └─ looplia registry → Manage skill registry sources                        │
│  └─ looplia skill    → Install and manage individual skills                 │
│  └─ looplia config   → Manage user profile                                  │
└─────────────────────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│  Workflow-as-Markdown + Per-Step Orchestration                              │
│  • Workflow definitions (workflows/*.md with YAML frontmatter)              │
│  • Each step: skill: + mission: syntax                                      │
│  • One step → one skill-executor call → multiple skills execution           │
│  • Validation-driven completion (validation.json)                           │
└─────────────────────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│  Two-Plugin Model + Skill Registry                                          │
│  ┌─────────────────────────────┬───────────────────────────────────────────┐│
│  │ looplia-core                │ looplia-writer                            ││
│  │ (Infrastructure)            │ (Domain)                                  ││
│  ├─────────────────────────────┼───────────────────────────────────────────┤│
│  │ • CLAUDE.md (entry point)   │ • workflows/writing-kit.md                ││
│  │ • commands/run.md, build.md │ • skills/media-reviewer/                  ││
│  │ • skills/workflow-executor/ │ • skills/idea-synthesis/                  ││
│  │ • skills/workflow-validator/│ • skills/writing-kit-assembler/           ││
│  │ • skills/registry-loader/   │ • user-profile.json                       ││
│  │ • hooks (TypeScript v0.7.4) │                                           ││
│  └─────────────────────────────┴───────────────────────────────────────────┘│
│  ┌─────────────────────────────────────────────────────────────────────────┐│
│  │ Skill Registry System (v0.7.0)                                          ││
│  │ • ~/.looplia/registry/skill-catalog.json (compiled from all sources)    ││
│  │ • ~/.looplia/plugins/ (third-party skills from marketplaces)            ││
│  │ • Default sources: Anthropic (42+ skills), ComposioHQ (29+ skills)      ││
│  └─────────────────────────────────────────────────────────────────────────┘│
└─────────────────────────────────────────────────────────────────────────────┘

Features

• Skills-First Architecture - Workflows declare skills + missions, not rigid agent definitions
• Skill Registry System - shadcn/ui-inspired registry for skill discovery and installation
• Selective Plugin Loading - Only load skills required by workflow (reduced context usage)
• Third-party Skills - Install community skills via git clone from marketplaces
• Marketplace Integration - 56+ skills installed by default (Anthropic + ComposioHQ)
• Per-Step Orchestration - One workflow step → one skill-executor call → multiple skills
• Workflow-as-Markdown - YAML frontmatter + markdown documentation
• Sandbox Isolation - Each run in sandbox/{id}/inputs/outputs/logs/
• Validation-Driven - Deterministic script-based output validation
• Two-Plugin Model - Separate infrastructure (core) from domain logic (writer)
• Streaming TUI - Real-time progress with tool execution display
• Smart Continuation - Resume workflows via --sandbox-id
• TypeScript - Full type safety with Zod schemas

Quick Start

# 1. Install dependencies
bun install

# 2. Build the project
bun run build

# 3. Initialize workspace (creates ~/.looplia/ with plugins)
bun run apps/cli/dist/index.js init --yes