Beehive

Versatile tools across Gemini CLI, Claude Code and Codex, with the intelligence to leverage & orchestrate them autonomously.

Important

Quick links to key resources:

• Setup guide
• Usage guide

Purpose

Agentic coding CLIs are fragmented: Claude Code, Gemini CLI, and Codex each have unique strengths but incompatible tooling.

Power users rotating between these (as is often the best option, given providers' unpredictable model throttling and capricious rate limit changes, even for paid plans) lose time rebuilding and reconfiguring context, tools, and custom workflows. At the same time, many (solo- or multi-agent) orchestration frameworks have considerable learning curves and enforce opinionated orchestration patterns (e.g. graphs, crews, pipelines), rather than adapting to users' ad-hoc workflows or permitting open-ended exploration/building.

This repo solves that by:

• Providing a unified, consistent set of 13 MCP servers (+ extra plugins) across all 3 platforms (Sourcegraph, Semgrep, Brave, Tavily, Context7, etc.)

• Defining 39 specialized agent roles once, usable everywhere (debugger, architect, security-compliance, etc.) and spawnable as cross-CLI subagents

• Minimizing task delegation overhead from ~5 minutes of setup to <30 seconds

• Reducing the learning curve to near-zero

  • Automated context injection ensures agents automatically and judiciously use all functionality made available by this repo
  • Minimal explicit directions needed from the user

Feature list

Consistent agent roles across 3 CLI platforms

• 39 specialized roles (architect, frontend, security, observability, debugger, data-eng, etc.)
• Same role definitions work everywhere via shared prompt sources
• Choose model per task (e.g. Claude for architecture, Gemini for broad code search)

2 ways of invoking agents

• As subagents (what you're probably more used to): spawn isolated tasks with separate context, get only the results back

  • Done using both:

    1. Claude Code's built-in subagents feature
    2. Zen's clink (for Codex and Gemini CLIs, as well as cross-model/-CLI collaboration)

• Direct use and interaction in main/current conversation (using custom-generated wrappers/commands):

  • Codex, Gemini CLIs: launch CLI with chosen agent active in main conversation
  • Claude Code: automatically activate any agent in the current conversation using slash commands

13 MCP servers

Including:

• Code search (Sourcegraph, Serena)
• Web research (Brave, Tavily, Fetch)
• API documentation (Context7)
• Memory/persistence (Qdrant, claude-mem)
• Security scanning (Semgrep)
• Browser automation (Playwright)

Automatic config injection

Meant to enable automatic use of features listed above by the CLIs (even when not using any particular agent).

All agents automatically read these files at startup:

• agents/reference/handoff-guidelines.md → when to delegate to subagents + which model to use

• agents/reference/compact-mcp-list.md → MCP tool selection guide

  • Serves as an entrypoint to documentation progressively disclosing each MCP servers' tool capabilities

Injected via these files (created in setup steps)

• ~/.claude/CLAUDE.md (Claude Code)
• ~/.gemini/GEMINI.md (Gemini CLI)
• ~/.codex/AGENTS.md (Codex)

with each of the 3 files above generated from templates (for portability regardless of repo clone location).

Note

CLI agent selection

• Setup scripts automatically detect which CLIs to configure based on user-scoped directory existence (~/.claude/, ~/.gemini/, ~/.codex/).
• More details can be found in SETUP.md.

Usage patterns

Subagents (for isolated execution)

Claude Code (Task tool):

"Have the architect subagent design this system"

Any CLI (Zen clink):

"clink with gemini architect to design API structure"
"clink with codex observability to analyze these metrics"

Direct use in the main conversation

Claude Code (via slash commands)

claude
> /explainer
# interactive conversation helping you understand the repo you spawned it in

Gemini/Codex (via wrapper scripts)

gemini-explainer                        # use explainer role w/ Gemini CLI
codex-architect --model gpt-5-codex     # architect role w/ GPT-5-Codex via Codex

Repo structure

agents/
├── claude-subagents/      # Claude Code Task tool definitions
├── clink-role-prompts/    # Zen clink role definitions (cross-CLI)
├── reference/             # MCP guides (injected via config files)
└── scripts/               # Setup automation

configs/
├── *.template             # Config templates with {{REPO_ROOT}} placeholders
└── scripts/               # Generates ~/.claude/CLAUDE.md, ~/.codex/AGENTS.md, etc.

tools/
├── tools.md               # Complete MCP listing
├── tools-decision-guide.md
└── scripts/               # MCP installation automation