npm for AI agents.
Browse, install, and publish the roles, skills, agents, memories, and integrations that power your AI workforce — strawhub.dev
strawhub install role ai-ceo
Resolving dependencies...
✓ role ai-ceo
✓ role pm
✓ skill project-planning
✓ skill task-breakdown
✓ role implementer
✓ skill git-workflow
✓ skill python-dev
✓ skill run-tests
✓ skill code-review
✓ role reviewer
✓ skill security-baseline
11 packages installed.
One command. Your entire AI company — skills, roles, and all their dependencies — ready to work.
AI agents are powerful. But every team reinvents the wheel:
- Write the same "code review" prompt for the tenth time
- Copy-paste skills between projects
- No way to share what works
StrawHub is a package registry for AI agent capabilities. Publish once, install anywhere.
pip install strawpot
strawhub install role ai-ceo
strawpot guiThat's it. StrawHub resolves every skill and sub-role your AI CEO needs. StrawPot runs them.
┌──────────────────────────────────────────────────────┐
│ StrawHub │
│ │
│ ┌─── Core ──────────────────────────────────────┐ │
│ │ Roles Skills │ │
│ │ ┌──────────┐ ┌──────────┐ │ │
│ │ │ai-ceo │ │git-wflow │ │ │
│ │ │pm │ │code-rev │ │ │
│ │ │implmtr │ │py-test │ │ │
│ │ │reviewer │ │debugging │ │ │
│ │ └──────────┘ └──────────┘ │ │
│ └───────────────────────────────────────────────┘ │
│ │
│ ┌─── Extensions ────────────────────────────────┐ │
│ │ Agents Memories Integrations │ │
│ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │
│ │ │claude │ │proj-ctx │ │telegram │ │ │
│ │ │codex │ │patterns │ │slack │ │ │
│ │ │gemini │ │decisions │ │discord │ │ │
│ │ └──────────┘ └──────────┘ └──────────┘ │ │
│ └───────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────┘
Job definitions that bundle skills. An ai-ceo depends on pm, implementer, and reviewer — StrawPot handles the delegation.
Atomic capabilities: writing code, reviewing PRs, running tests, searching docs. Roles compose skills; skills are the building blocks.
Runtime wrappers for Claude Code, Codex, Gemini, or your own CLI.
Persistent knowledge banks: project context, code patterns, past decisions.
Adapters that bridge StrawPot to external services like Telegram, Slack, or Discord.
# ai-ceo/ROLE.md
---
name: ai-ceo
description: "Orchestrator that analyzes tasks, discovers all installed roles,
and delegates to the best-fit role."
metadata:
strawpot:
dependencies:
roles:
- "*"
default_agent: strawpot-claude-code
---
# AI CEO
You are a routing layer with judgment. The user brings you a task —
you figure out which role on your team should handle it, write a clear
task description, and delegate. That's the entire job.No Python. No YAML config files. No orchestration code. One Markdown file per role.
# CrewAI: 40 lines of Python to define a 2-agent team
# OpenClaw: Skills in Markdown, but no roles or dependency resolution
# StrawPot: One ROLE.md file. Done.
| CrewAI | OpenClaw | StrawPot + StrawHub | |
|---|---|---|---|
| Format | YAML + Python | JSON5 + Markdown | Markdown only |
| Skills / Tools | Python (tools) | Markdown (skills) | Markdown (skills) |
| Roles | Agent attribute | — | Standalone Markdown |
| Memory | Python config | Markdown/YAML (local) | Markdown (installable) |
| Skill dependency resolution | — | — | Automatic |
| Multi-agent delegation | Python config | Runtime (subagent spawn) | Declarative (role deps) |
| Package registry | — | — | StrawHub |
Install a role. Everything it needs comes with it.
strawhub install role ai-ceo
ai-ceo
├─ role: pm
│ ├─ skill: project-planning
│ └─ skill: task-breakdown
├─ role: implementer
│ ├─ skill: git-workflow
│ ├─ skill: python-dev
│ ├─ skill: run-tests
│ └─ skill: code-review
└─ role: reviewer
├─ skill: code-review (already installed)
└─ skill: security-baseline
Skills use client-side DFS. Roles use server-side topological sort with cycle detection. You don't have to think about any of it.
# Install
strawhub install role ai-ceo
strawhub install skill git-workflow
strawhub install agent strawpot-claude-code
strawhub install memory project-context
strawhub install integration telegram
# Discover
strawhub search "code review"
strawhub info role ai-ceo
strawhub list
# Validate & Publish
strawhub validate skill ./my-skill/
strawhub publish role ./my-role/
strawhub publish skill ./my-skill/
# Manage
strawhub login
strawhub whoami| Project | What it does |
|---|---|
| StrawPot | Runtime — runs role-based AI agents locally |
| StrawHub | Registry — distributes roles, skills, agents, memories, and integrations |
| Denden | Transport — gRPC bridge between agents and the orchestrator |
User task → StrawPot runtime → Role (ai-ceo)
├─ Sub-role (pm)
│ └─ Skills (project-planning, task-breakdown)
├─ Sub-role (implementer)
│ ├─ Skills (git-workflow, python-dev)
│ └─ Agent (strawpot-claude-code)
└─ Sub-role (reviewer)
├─ Skills (code-review, security-baseline)
└─ Agent (gemini)
Great engineers shouldn't re-invent agent skills.
They should install them.
strawhub.dev