by stephenleo
🎮 Codex Micro functionality using any Gaming Controller on any Coding Harness!
# Add to your Claude Code skills
git clone https://github.com/stephenleo/OpenMicroOpenMicro is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by stephenleo. 🎮 Codex Micro functionality using any Gaming Controller on any Coding Harness!. It has 51 GitHub stars.
OpenMicro's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/stephenleo/OpenMicro" and add it to your Claude Code skills directory (see the Installation section above).
OpenMicro is primarily written in TypeScript. It is open-source under stephenleo on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh OpenMicro against similar tools.
No comments yet. Be the first to share your thoughts!
Unlocks once the catalog security scan passes (runs nightly).
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
Bring Codex Micro to any gaming controller and coding harness
You need macOS, Node.js 22 or newer, Claude Code or Codex CLI, and a connected controller. OpenMicro is macOS-first; other platforms are not yet tested.
npm i -g openmicro
openmicro claude # or just: openmicro
openmicro codex
OpenMicro installs its lifecycle hooks automatically. If Codex reports that its hooks changed, open /hooks in Codex and trust the OpenMicro hooks.
Controller support depends on the exact device and connection. Check the controller compatibility guide before you start, or run openmicro doctor to test your controller.

| Control | Action |
|---|---|
| south (✕ / A) | Submit or confirm |
| east (○ / B) | Interrupt or dismiss |
| north (△ / Y) | Push-to-talk |
| west (□ / X) | Start a new chat |
| d-pad | Navigate TUI menus; repeats while held |
| left stick flick up / down / left / right | Review PR / debug / refactor / write tests |
| right stick rotate clockwise / counterclockwise | Increase / decrease thinking depth |
| R1 | Cycle modes (Shift+Tab) |
| R2 | Clear the input line (Ctrl+U) |
| touchpad click | Focus the next session by default, where supported |
| L2 | Cycle herdr spaces; touchpad then cycles agents in the space |
Stick flicks fire after returning to center; each quarter-turn steps thinking depth once. Hold L1 with south, east, west, north, d-pad up, or d-pad down to select one of six layers. The first layer ships with these defaults; the other five start empty. Other controls are unbound by default and remappable.
Voice and thinking-depth support varies by harness; see OpenMicro feature parity.
✅ supported · ⚠️ setup required or best-effort · — no verified equivalent
| Capability | Claude Code | Codex CLI |
|---|---|---|
| Launch and forward CLI arguments | ✅ openmicro claude |
✅ openmicro codex |
| Submit / confirm | ✅ Enter | ✅ Enter |
| Interrupt / dismiss | ✅ Escape | ✅ Escape |
| Start a new chat | ✅ /clear |
✅ /new |
| D-pad TUI navigation | ✅ Arrow-key passthrough | ✅ Arrow-key passthrough |
| Stick-triggered workflow prompts | ✅ Supported | ✅ Supported |
| Push-to-talk | ✅ Enable with /voice |
— No equivalent |
| Thinking-depth dial | ✅ /effort, low → max |
— No deterministic binding |
| Six remappable control layers | ✅ Supported | ✅ Supported |
| Multi-session focus switching | ✅ Supported | ✅ Supported |
| Executing status | ✅ Prompt and tool hooks | ✅ Prompt and tool hooks |
| Waiting-for-input status | ✅ Questions and notifications | ✅ Permission requests |
| Stop status | ✅ Stop hook | ✅ Stop hook |
| Error status | ⚠️ Notification-text matching | — No error hook signal |
| Hook installation | ✅ Automatic | ⚠️ Trust changes in /hooks |
Layers, workflows, navigation, and session switching are handled by OpenMicro itself. Harness-specific gaps are left unmapped instead of sending guessed keystrokes. A Stop event means the agent stopped; it does not guarantee success.
OpenMicro creates ~/.openmicro/config.json on first run. Edit bindings, layer colors, and workflow prompt text there. Invalid configuration stops startup without overwriting the file.
{
"layers": [
{
"name": "Layer 1",
"color": { "r": 255, "g": 255, "b": 255 },
"bindings": {
"south": { "type": "accept" },
"lstick_up": { "type": "workflow", "presetId": "review-pr" },
"rstick_cw": { "type": "thinking_depth", "delta": 1 }
}
},
{ "name": "Layer 2", "color": { "r": 160, "g": 32, "b": 240 }, "bindings": {} },
{ "name": "Layer 3", "color": { "r": 0, "g": 255, "b": 255 }, "bindings": {} },
{ "name": "Layer 4", "color": { "r": 255, "g": 140, "b": 0 }, "bindings": {} },
{ "name": "Layer 5", "color": { "r": 255, "g": 20, "b": 147 }, "bindings": {} },
{ "name": "Layer 6", "color": { "r": 255, "g": 255, "b": 0 }, "bindings": {} }
],
"workflows": {
"review-pr": "Review this PR for correctness, security, and style issues."
}
}
Binding keys can be buttons such as south and dpad_up, or gestures such as lstick_up and rstick_cw. Actions include accept, reject, push_to_talk, new_chat, thinking_depth, workflow, prompt, focus_session, layer, herdr_space, and raw keys.
The first OpenMicro process owns the controller and becomes the host. Later processes register as clients, so one controller can drive several terminal sessions. On supported pads, touchpad click cycles focus by default.
On DualSense, the lightbar follows the focused session: blue while executing, amber while waiting, green when stopped, red on a detected error, and dim white while idle. The five player LEDs show occupied session slots.
OpenMicro treats herdr as a first-class environment. Everything below is automatic and a no-op outside herdr or when the herdr CLI is absent.
openmicro doctor
The diagnostic checks controller input and, on DualSense, lightbar/player-LED output. It writes a <vid>-<pid>-<transport>.json report that can be added unchanged to test/fixtures/controllers/; CI then replays the captured inputs as a regression test. See CONTROLLERS.md for supported devices, connection-specific notes, and contribution steps.
The public openmicro/harness API exposes the Harness contract and registerHarness(). Implement the contract, return null for actions without a verified CLI equivalent, and register it before OpenMicro resolves the harness.
import { registerHarness } from 'openmicro/harness'
import type { Harness } from 'openmicro/harness'
const myHarness: Harness = {
kind: 'my-cli',
command: 'my-cli',
buildArgs: (args) => args,
installHooks: () => ({ changed: false, trustNotice: null }),
stateForHookEvent: () => null,
resolveAction: (action) => {
if (action.type === 'accept') return { bytes: '\r' }
if (action.type === 'reject') return { bytes: '\x1b' }
if (action.type === 'prompt') return { bytes: `${action.text}\r` }
if (action.type === 'keys') return { bytes: action.bytes }
return null
},
}
registerHarness(myHarness)
The binary does not load harness plugins from configuration yet, so a third-party registration currently needs a small custom entry point.