agentleFS
EskyeeEskyee/agentbot-opensource/skills/codegrid/SKILL.md

codegrid

Skill17 starsChanged 5 months agoMITView on GitHub

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, monitor, debate). The differentiator: multiple coding agents collaborating on one canvas, addressable by stable session_id, with a read → message → read protocol built for orchestration.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

CodeGrid — agent operating manual

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.

When to load which reference

You're about to…Load
Discover what's running, open a folder, spawn a session, or drive CodeGrid from outsideusing-codegrid
Hand work to / consult / review with another agent already running in CodeGridcodegrid-agent-bus
BothLoad using-codegrid first, then codegrid-agent-bus

Quickstart (inside a CodeGrid pane)

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.

Quickstart (from outside CodeGrid, e.g. another tool)

SOCK="$(cat ~/.codegrid/socket-path 2>/dev/null || echo ~/.codegrid/socket)"
printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"agent_list"}' | nc -U "$SOCK"

# Or deep-link the installed app:
open "codegrid://open?path=/abs/path/to/repo&type=codex"
open "codegrid://new"

Install

install the codegrid skill from https://github.com/BankrBot/skills/tree/main/codegrid

Source

  • Canonical source of truth lives in the CodeGrid repo under skills/ (skills/using-codegrid/SKILL.md, skills/codegrid-agent-bus/SKILL.md). This Bankr Skills provider is a published copy kept in sync.
  • App download + full docs: https://codegrid.app.
  • Token (Base): 0x6B456E66524aEC1792013eF9DFE87e3F84311ba3 — see https://codegrid.app/token.

Safety defaults an agent should respect

  • Read before you write. read_pane is free; never message a busy agent.
  • Address by session_id, never pane number — pane numbers shift.
  • Be explicit about scope. Other agents may be in YOLO/autonomous mode and will act on whatever you send. Say "propose only, do not edit files" when you want analysis.
  • One message, then wait. Multiple messages interleave into the target's input box and corrupt it.
  • Bound debates and don't spawn unbounded helpers. Converge and report to the user.
  • Stay local. Everything is same-machine IPC; there is no remote bus.

Full guidance: see the two reference files.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.