Skip to content
Back to skills

Omk Neo

ASecurity

Entry point for Neo computer-use work in this checkout. Route browser, website, research, code-review and MCP-setup tasks to the matching Neo skill after checking actual capabilities, and explain Neo skill/MCP availability without claiming connections.

  • 145 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 19, 2026
ai-agentscode-reviewapiperformancedocumentation

Works with

  • api
  • mcp

Security analysis

A100/100

Scanned September 23, 2026

npx -y skills add dmae97/omk --skill omk-neo --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Omk Neo?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Omk Neo
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/dmae97-omk-neo/badge)](https://www.skillsdirectory.com/skills/dmae97-omk-neo)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: omk-neo
description: Entry point for Neo computer-use work in this checkout. Route browser, website, research, code-review and MCP-setup tasks to the matching Neo skill after checking actual capabilities, and explain Neo skill/MCP availability without claiming connections.
---

# Neo router

Start here for Neo tasks. This skill routes; it does not itself drive browsers, write sites, or configure servers. The six public Neo skills shipped in coding-agent 1.2.0 under `resources/neo/skills`. This checkout may still differ from the published package, so verify `omk neo list` before recommending commands to others.

## Route by target

Neo workflow skills are bundled with the installed package under `resources/neo/skills` and are discovered by skill name. Load by name, for example with `!skill:omk-browser` or `/skill:omk-browser`.

| Task | Load skill |
| --- | --- |
| Desktop app, native GUI, or model/capability questions | `omk-computeruse` |
| Website interaction, screenshots, structured extraction | `omk-browser` |
| Building or validating a site in this repository | `omk-site` |
| Current technical documentation lookup | `omk-research` |
| Reviewing a change before handoff | `omk-code-review` |
| Installing or connecting Neo MCP presets | `omk-mcp-setup` |

Read only the matched skill, plus a second one when the task genuinely crosses domains (for example site work needs browser verification). Load `omk-computeruse` when the target is native desktop or you must first decide which model capabilities the task needs. If a matched name is absent from your session inventory, check `omk neo list` for packaging state instead of inventing a path; a missing name is a packaging problem, not a reason to fake a workflow.

## Check Neo state before acting

1. `omk neo list` shows packaged skills and offered presets from the built installation. `connected: null` is unprobed, not zero failures.
2. Bundled skills fill only names that user, project, or explicit skills do not already own. A same-name project skill always wins.
3. This working tree's bundle lives at `packages/coding-agent/resources/neo/skills`. The local build also copies it to `dist/resources`; the published package verification remains a separate release task.
4. Restart OMK after a rebuild; an active session does not hot-reload source.

## MCP discipline

MCP setup is opt-in and scoped. `omk neo setup playwright context7` previews the exact target and disabled entries without writing; `--apply` needs explicit approval for the target file and network effects. Never overwrite an existing `.omk/mcp.json`; use `omk neo mcp-config` and a reviewed manual merge instead. Sandbox launch failure must be diagnosed, never bypassed with `--no-sandbox`. Profile isolation is not OS or network isolation.

Current machine status, checked 2026-09-17: a Playwright MCP connection answered a read-only probe; Context7 returned an invalid API-key error. Treat those as per-host facts to recheck, not standing connectivity claims. Do not modify credentials without an explicit request.

## Report

End with runtime, target, actions, verification evidence, and `VERIFIED`, `PARTIAL`, or `BLOCKED`. Distinguish packaged, loaded, configured, connected, and verified states. Never claim deployment, publishing, or model-performance gains that no evidence in this checkout supports.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.