nexus-windows-codex
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| NEXUS_GARDE_FOUS | No | Path to an alternative guardrails policy file. If missing or invalid, every action is refused. | |
| NEXUS_JOURNAL_HOME | No | Overrides the directory where the short monthly JSONL journal is written. | |
| NEXUS_WINDOWS_INDICATOR | No | Set to 0 to disable the on-screen busy indicator shown while actions run. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| windowsA | Application windows on the desktop (id, app, title, pid), like the taskbar, plus open dialogs (a file "Open" box…) with dialogOf: the window that owns them; look at a dialog by its id or title. Password managers are never listed. filter: part of a title or app. |
| lookA | See a window in the background, without bringing it forward: UI Automation tree (roles, names, ids, values, states) and, with mode image or both, a native capture that works even when the window is covered (WebP, at most 2000 px; unchanged since the last image sent: not attached again, mode image always attaches it). A window with almost no tree (Tk, canvas, game: fewer than 6 elements) also gets the text of its image read by OCR (ocr.lines: text, point, screenshotId), clicked with point + screenshotId. Selects that window for act; an observation serves one act and expires after 60 s. Returns its observation id: pass it to act (required from protocol 2026-07-28). |
| actA | Steps on the window of the last look, in the background first (Win32 messages, UI Automation, DevTools for Chromium), the real mouse/keyboard only as a fallback under the shared foreground token. One step, or up to 32 each with its own expect; stops at the first step not verified. Per step: ✓ verified, = already done, ~ unknown (never replayed: look again), ✗ failed or refused. interaction: true keeps the window in front for the whole call (shared foreground token, at most 30 s, given back at the end) for a menu that closes when the foreground goes back (a context menu opened by a real right click, then its command): selectors only, every step with its expect; look again afterwards. |
| launchA | Open an application (name, absolute path or Store app id, with its arguments), a folder or a document (absolute path) or a Settings page (ms-settings:...) and give the foreground back to the user. Returns the windows attributed to this launch, each with its proof: "process" (the process the shell returned), "folder" (a new Explorer window showing exactly that folder) or "application" (the requested application, started after the request: another instance started at the same moment could be taken for it; a Store app by its app id); other new windows come back as candidates (outcome unknown). Look at one before acting. |
| windowA | close: close the window (one close message, read back; a "save?" prompt is reported, never answered). |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 5 tools
The core purposes (enumerate, observe, act, launch, close) are conceptually distinct, but `windows` (list all desktop windows) and `window` (close a window) differ by a single letter and are easy to misselect when an agent scans tool names. The `look`/`act` pairing is intentionally coupled (act requires the last look's observation id), which is documented but adds another boundary the agent must track.
Naming mixes bare verbs (`look`, `act`, `launch`) with bare nouns (`windows`, `window`) and no verb_noun pattern, so there is no predictable rule to infer a tool's action from its name. The near-homograph `window`/`windows` compounds the inconsistency.
Five tools map cleanly onto the workflow: enumerate windows, observe one, act on it, launch apps, close a window. Each tool earns its place and there is no redundancy or filler.
Observation, interaction, launching, and closing are covered, but window lifecycle mutations beyond close are absent: no move, resize, minimize/maximize/restore, or explicit activate/focus operation. These are notable gaps for a server whose domain is desktop window management, forcing agents to fall back on UI-level workarounds.