Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →bankrbot/codegrid
CodeGrid is a native macOS canvas where multiple coding agents (Claude, Codex, Gemini, Cursor, Grok, shells) run side by side in panes and collaborate via a local agent bus — no tmux, no cloud, no account, no stored API keys. Install this skill when an agent should know how to operate inside a CodeGrid pane, drive the workspace from outside (control socket or codegrid:// deep links), spawn or message sibling agents, or coordinate multi-agent work (delegate, review, pipeline, parallel fan-out, mo
Passed 24 of 25 test cases with the skill, and 9 of 25 without it.
and 17 more in the full report
15 fixed9 passed either way1 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 15 test cases the skill fixed. Same model, same prompt, with and without the skill.
Before dispatching a task to another worker pane in CodeGrid, what step should an agent take to ensure it does not interrupt an ongoing execution?
| Checked | Without skill | With skill |
|---|---|---|
| Recommends calling read_pane to check target status before sending a request. | Missed | Passed |
The instructions your model reads when the skill loads.
CodeGrid runs many coding agents side by side on an infinite 2D canvas. Each
pane is one process (claude, codex, gemini, cursor-agent, grok, or a
shell) with its own working directory, git branch, and stable session_id. The
agent bus lets any pane discover, read, and message any other pane locally.
This skill is the umbrella entry point. Two references cover the full surface:
Operating CodeGrid — the mental model
(canvas / pane / workspace / agent / bus), MCP tools (list_agents,
read_pane, message_agent), the control-socket JSON-RPC API
(agent_list, agent_read, agent_send, open_folder, new_session,
new_workspace), the codegrid:// deep-link scheme, and an operating
playbook with recipes.
Agent-bus collaboration — the
read → message → read protocol, identifying agents by role and
session_id, orchestration patterns (delegate, review, pipeline, parallel
fan-out, monitor, debate), etiquette and scope safety, loop/runaway
prevention, failure recovery, and worked end-to-end examples.
| You're about to… | Load |
|---|---|
| Discover what's running, open a folder, spawn a session, or drive CodeGrid from outside | using-codegrid |
| Hand work to / consult / review with another agent already running in CodeGrid | codegrid-agent-bus |
| Both | Load using-codegrid first, then codegrid-agent-bus |
list_agents() # discover panes (session_id, role, status)
read_pane(<session_id>) # see what they're doing — always safe
message_agent(<session_id>, "[from <you>] <self-contained request>")
… wait, then read_pane(<session_id>) for the reply.
test cases passed, vs 9 / 25 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.