Drop-in · Agent Skills

Teach any agent to use CodeGrid.

Skills are small, self-contained SKILL.md packages that give an AI agent new abilities. Ours teach Claude, Codex, Gemini, Cursor, and Grok how to drive CodeGrid and collaborate with each other — drop them into Claude Code, any AGENTS.md agent, or a Bankr agent via OpenClaw.

Get the skills on GitHub ↓Read the docs →
The operating manual
using-codegrid

Teaches an agent to drive CodeGrid — discover what's running, read and message sibling panes, open projects, and control the workspace over the local socket or codegrid:// deep links.

  • ›Discover, spawn, and address panes by stable session_id
  • ›The full control-socket JSON-RPC reference (agent_list / read / send, open_folder, new_session)
  • ›The codegrid:// deep-link scheme for "Open in CodeGrid" buttons
  • ›The canvas / pane / workspace mental model + operating playbook

Use when an agent is running inside a CodeGrid pane, or the user asks to spawn/list/control agents or drive the workspace.

View SKILL.md →
The collaboration protocol
codegrid-agent-bus

Teaches an agent to work with other agents through the Agent Bus — delegate, review, run a pipeline, fan out work, or get a second opinion, without talking over a busy agent.

  • ›The read → message → read protocol and agent etiquette
  • ›Orchestration patterns: delegate, review, pipeline, parallel fan-out, monitor, debate
  • ›Loop & runaway prevention, scope safety, failure recovery
  • ›Worked end-to-end examples across Claude, Codex, Gemini, Cursor, Grok

Use when the user says "ask Codex to…", "have Gemini review…", "split this between the agents", or any cross-agent coordination.

View SKILL.md →

What's a skill, exactly?

A skill is a single Markdown file — YAML frontmatter plus a body — that an agent reads to learn a capability. No install step, no dependency, no runtime: just context, written for agents.

Because the format is shared across Claude Code, AGENTS.md-style agents, and Bankr's OpenClaw registry, one skill works everywhere. Write it once; every agent that reads it gets smarter about CodeGrid.

Install in three ways

01
Claude Code

Copy a skill folder into your skills directory — Claude loads it automatically.

git clone --depth 1 \
  https://github.com/ZipLyne-Agency/\
  CodeGrid-Claude-Code-Terminal
cp -r CodeGrid-*/skills/* \
  ~/.claude/skills/
02
Any agent (AGENTS.md)

Paste a SKILL.md body into your agent's instruction file — Codex, Gemini, Cursor, Grok, and more.

# open the skill, copy the body into
# your AGENTS.md / system prompt
open skills/using-codegrid/SKILL.md
03
Bankr · OpenClaw

Point OpenClaw at the repo and pick the skill — your Bankr agent learns CodeGrid in one line.

install the codegrid skill from
https://github.com/ZipLyne-Agency/
CodeGrid-Claude-Code-Terminal/tree/
main/skills

Both skills assume the CodeGrid app is running — the Agent Bus and control socket are local, same-machine IPC.

Run the agents. Then let them run together.

Install CodeGrid, drop in the skills, and your agents stop working in isolation — they discover, delegate to, and review each other on one canvas.

Download for Mac ↓How the Agent Bus works →