hive

Hive

Give your AI coding assistant superpowers. 60+ skills, 10 agents, LLM handlers for 6 providers — works with Claude Code, Cursor, Gemini CLI, and Codex.

What is Hive?

Hive turns your AI assistant into a full dev team. Instead of copy-pasting API docs or writing boilerplate, use slash commands:

/shape                                       # create a well-defined issue
/dev                                         # orchestrate work across agents
/ship_it                                     # review, fix, merge, deploy, verify
/perplexity "best auth library for Next.js"  # search the web
/gmail inbox --unread                        # check email
/generate_image "a mountain landscape"       # create an image

Need something bigger? Delegate to specialist agents that handle design, frontend, backend, QA, SEO, and more.

Quick Start

# 1. Clone into your workspace
cd ~/workspace
git clone https://github.com/Skywalking-dev/hive.git

# 2. Install dependencies
cd hive
uv sync

# 3. Configure API keys (add only the ones you need)
cp .env.example .env
# Edit .env with your keys

# 4. Install skill packs
uv run hive install --all

# 5. Setup workspace
uv run hive setup

That's it. Open Claude Code from ~/workspace/ and all skills, agents, and handlers are available.

What hive setup does

[hive] Actions:
  Symlink ~/workspace/.claude/skills  -> hive/skills     # skills available
  Symlink ~/workspace/.claude/agents  -> hive/agents     # agents available
  Generate .cursor/rules/ from skills                     # Cursor compatible
  Merge .mcp.json configs                                 # MCP servers
  Merge hive/CLAUDE.md into workspace CLAUDE.md           # handler docs
  Verify API keys in hive/.env                            # missing key checklist

Setup is idempotent — run it again after pulling updates to sync everything.

Workspace structure after setup

~/workspace/                          <- open your AI tool here
├── .claude/
│   ├── skills/ -> hive/skills/       <- symlink
│   └── agents/ -> hive/agents/       <- symlink
├── .cursor/rules/                    <- generated .mdc files
├── .mcp.json                         <- merged MCP servers
├── CLAUDE.md                         <- workspace config + hive handler docs
├── hive/                             <- this repo (source of truth)
│   ├── skills/
│   ├── agents/
│   ├── scripts/
│   └── .env                          <- API keys (gitignored)
└── your-projects/

LLM Handlers

Hive includes shell handlers for 6 LLM providers. All are OAI-compatible and load keys from hive/.env.

Handler	Default Model	Cost out/MTok	Commands
groq_handler.sh	Llama 4 Scout	Free	ask, models
deepseek_handler.sh	DeepSeek V4	$0.50	ask, models
openrouter_handler.sh	deepseek-chat	Free with --free	ask, models, free
gemini_handler.sh	Gemini 2.5 Flash	$2.50 (free tier)	ask, search, embed
openai_handler.sh	GPT-4.1	$8.00	ask, responses, embeddings
perplexity_handler.sh	Sonar Pro	$15.00	ask, search, agent

Usage

# Free and fast
hive/scripts/groq_handler.sh ask "Explain this error" --json

# Cheapest frontier model
hive/scripts/deepseek_handler.sh ask "Review this code" --system "You are a code reviewer"

# Deep reasoning
hive/scripts/deepseek_handler.sh ask "Analyze this architecture" --reasoner

# Free via OpenRouter
hive/scripts/openrouter_handler.sh ask "Summarize this" --free

# With fallback chain
hive/scripts/openrouter_handler.sh ask "Translate this" --fallbacks "google/gemini-2.5-flash,meta-llama/llama-4-scout"

# Web search with citations
hive/scripts/perplexity_handler.sh search "latest Next.js release"

Model routing by task

Task	Best pick	Why
Classification / routing	Groq	Free, 3000+ tok/s
Reviews / batch / bulk	DeepSeek V4	Cheapest frontier ($0.50/MTok)
Complex reasoning	DeepSeek R1	91% cheaper than Opus
Code generation	OpenAI or DeepSeek	Dedicated code models
Web search / grounded	Perplexity Sonar	Built-in search + citations
Vision / multimodal	Gemini Flash	Free tier, 1M context
Fallback / free models	OpenRouter --free	29+ free models

Skills

Core (always installed)

Command	What it does
/shape	Guided issue creation with discovery
/capture	Quick idea to backlog
/refine	Technical breakdown into agent sub-issues
/dev	Orchestrate work across agents
/push_it	Commit + push + open PR
/ship_it	Full pipeline: review, fix, merge, deploy, verify
/pr-review	PR review with Linear sync
/reunion	Multi-agent meeting

Senses (sensory extensions)