The AgntUX orchestrator. Triages action items and queries your knowledge store.
Installs first; everything else hangs off it. Maintains your `<agntux project root>/` knowledge store, triages action items per your `user.md` preferences, and keeps `_index.md` files in sync as ingest plugins write new data.
Screenshots

Action item triage
The primary inline UI. Priority-sorted open action items with snooze / dismiss / done buttons and per-item suggested-action buttons that route into source plugins.
About this plugin
AgntUX Core
The AgntUX orchestrator. Triages action items and queries your knowledge store.
What it does
AgntUX Core is the foundation plugin that all other AgntUX plugins build upon. It maintains your knowledge store, triages action items according to your preferences, and coordinates between ingest plugins to keep your data fresh and organized.
Install
Install AgntUX Core first before installing any other AgntUX plugin. It provides the shared knowledge store and orchestration layer that other plugins depend on.
After installing, run /agntux onboard once to create your <agntux project root>/user.md profile
and bootstrap the tenant schema.
Quickstart
agntux-core ships a single entry-point skill — /agntux. Sub-commands route
to the right behaviour by reading the first token of $ARGUMENTS; bare
/agntux defaults to ask-mode and infers from the natural-language prompt.
| Command | Purpose |
|---|---|
/agntux onboard | First-run interview + schema bootstrap. Run once. Re-runs walk newly-installed plugins. |
/agntux profile | Edit preferences, glossary, identity, sources. |
/agntux teach {plugin-slug} | Capture per-plugin rules ("never raise email from X"). |
/agntux schema [review|edit] [plugin-slug] | Review or edit the tenant schema. |
/agntux sync {plugin-slug} | Manually trigger an ingest pass for an installed plugin (bare names like slack expand). |
/agntux ask "..." | Catch-all for natural-language queries and inline status edits. |
/agntux feedback-review | Background pattern detection over resolved actions (scheduled task target). |
/agntux triage-digest | Daily text-digest fallback when the interactive triage UI doesn't render (scheduled task target). |
You can also speak naturally — Claude auto-dispatches to /agntux from
its description, and the router infers the right sub-command from the
prompt (e.g. saying "show triage" or "what's hot" invokes the
interactive triage UI directly via the agntux_core_triage_view MCP
tool — no skill in the loop).
Recommended scheduled tasks
| Task | Prompt body | Cadence |
|---|---|---|
| Daily action-item digest | /agntux triage-digest | Daily 08:00 |
| Daily feedback review | /agntux feedback-review | Daily 16:00 |
UI
agntux-core renders one MCP App: ui://triage. Saying "what's hot",
"show triage", or "what should I look at" triggers the host's tool
selector to invoke agntux_core_triage_view directly — that tool's
description carries the trigger phrases. The UI shows priority-sorted
open action items with inline mutation controls and per-item
suggested-action buttons that route into source plugins via
sendFollowUpMessage. The component runs server-side reads against
<agntux project root>/actions/; arguments to the view tool are
zero-required so the LLM spends ~no tokens on tool args.
For scheduled-background fires (Daily 08:00 by default), the
/agntux triage-digest sub-command emits a text digest via the same
data source — no UI, no audience required.
The previous entity-browser UI handler was retired in 5.0.0. Entity
navigation now goes through /agntux ask (e.g. "tell me about
person/avery-rivera").
Configuration
Configure your preferences in <agntux project root>/user.md. This
file controls how the orchestrator prioritizes action items and manages
your workflow. Run /agntux profile to edit it.
Limitations
- Requires at least one ingest plugin to populate the knowledge store with real data.
- Knowledge store lives on your local machine; no cloud sync at MVP.
Hooks
This plugin ships a small set of plugin-specific hooks under hooks/ for
schema and index validation:
hooks/validate-schema.mjs— PreToolUse, blocks malformed entity writes.hooks/validate-contract.mjs— PreToolUse, blocks contract violations.hooks/maintain-index.mjs— PostToolUse, keeps the entity index current.
License
Apache License 2.0. See the LICENSE and NOTICE files at the repo
root for full terms.
Support
- Bugs and proposals: https://github.com/AgntUX/AUX-plugins/issues
- License: see the LICENSE file at the repo root.
What's new
v10.1.02026-05-25Added
- New local MCP tool
agntux_core_create_project_directory. Creates the AgntUX project root at~/agntux(no-op if it already exists) and returns the absolute path. Takes no arguments — the target is always~/agntux, so the tool can't be used as a general-purposemkdirprimitive. Lives on the HOME-scoped local stdio server alongsideagntux_core_sync_installed_plugins.
Changed
- Onboarding Stage 0 no longer asks the user to run a terminal command
to create the
agntuxfolder./agntux onboardnow callsagntux_core_create_project_directoryto create~/agntux, then passes the returned absolute path to the host'srequest_cowork_directorytool to obtain working permission.
Why
In Claude Cowork the old flow was a dead end: Cowork's security guardrail
won't let the agent create a directory until it has permission to work in
the parent, and it isn't allowed to request permission for the user's home
directory — so ~/agntux could never be created interactively. The local
MCP server runs outside that cwd guardrail (the same HOME-scoped seam
sync-installed-plugins uses), so it can create ~/agntux directly, and
onboarding then requests access to the now-existing folder.
Supported prompts
Paste these into a scheduled task or send them to the AI host directly.
/agntuxOne entry point for AgntUX. Sub-commands: onboard, profile, schema, teach {slug}, sync {slug}, ask {…}, feedback-review, triage-digest. Bare /agntux defaults to ask-mode.
Install
/plugin install agntux-core@agntux