Skills, commands, agents & plugins

Five manager widgets let you browse, create, and edit every reusable Claude Code artifact — folder-form skills, slash commands, agent definitions, plugins, and text snippets — without leaving the browser.

All five share the same shape: a search box that matches names, descriptions, and body text; scope filter pills with live counts; and expandable cards. The file-backed managers (skills, commands, agents) add a grid/list view toggle, a "this project" switch, and an inline raw-Markdown editor on every editable card. Saves are mtime-guarded — if the file changed on disk since you opened it, the save fails loudly instead of silently clobbering the other edit.

Scopes: project, personal, plugin

Every artifact lives in one of three scopes, and each widget groups its cards accordingly:

Scope	Where it lives	Notes
Project	<project>/.claude/ (skills/, commands/, agents/)	Ships with the repo; visible to everyone who clones it
Personal	~/.claude/	Follows you across projects on this server
Plugin	Installed plugins	Read-only — edit by duplicating into project or personal scope

The this project toggle hides everything except project-scope artifacts, and the pills filter one scope at a time. Right-click any card for a context menu with Copy path, Duplicate to Project / Personal (also how you make an editable copy of a read-only plugin artifact), and Delete where the file is editable.

Invoking artifacts from chat

Each artifact type has a trigger character in the message box:

• / at the start of the input — commands and skills, autocompleted from the whole catalog. Skills are invoked as /skill-name too.
• # — snippets and agents. Selecting a snippet inserts its text; selecting an agent inserts an invocation phrase.
• $ anywhere in the prompt — skills picker, which inserts the skill's /slash-name invocation into your sentence.

The pickers, their keyboard behavior, and the @ file picker are covered in Writing prompts & commands; the full command catalog is in the command reference.

Skills manager

Open with Alt+K. Lists folder-form skills — directories containing a SKILL.md — from all three scopes.

Each card shows the skill's /name, its description, and meta chips: tool count (from the allowed-tools frontmatter), supporting-file count, and the invocation mode (auto + manual, manual only, auto only, or disabled). Badges mark plugin skills, read-only files, and skills that run in a forked subagent (context: fork). A ⚠ marker appears on a skill that is shadowed by another skill with the same name in a higher-priority scope.

Click a card to expand the inline editor for the raw SKILL.md — Save stays disabled until you change something, and Open in full editor hands the file to the file-preview widget for a proper CodeMirror session. Plugin skills open the same view read-only.

New Skill opens a creation dialog: pick a scope (project or personal), a name (lowercase letters, numbers, hyphens — it becomes the /slash-command), a description ("Claude uses this to decide when to auto-invoke"), and a starting template:

• Blank — empty SKILL.md
• Task workflow — manual-invocation workflow with numbered steps
• Reference content — standing conventions Claude auto-applies
• Subagent fork — runs in an isolated Explore subagent
• Visual output — bundles a script that generates visual output

The header has two extra buttons: a wrench that jumps to the Plugins manager and a refresh that re-scans the disk.

Commands manager

Open with Alt+Shift+K. This is the full slash-command catalog — everything the / picker can complete — grouped into five origins:

• Built-in (Claude CLI) — the CLI's own commands plus the bridge's client-side commands (/help, /fork, …). Read-only.
• Project — legacy single-file commands at .claude/commands/<name>.md in the project.
• Personal — the same, in ~/.claude/commands/.
• Plugin — commands shipped by installed plugins. Read-only.
• Custom (saved on this device) — prompt and shell commands you saved in the browser. Shown here read-only; they are managed where you created them.

Project and personal command files get the same inline editor as skills. Right-click an editable file command for two extra actions: Upgrade to skill folder, which converts the flat .md into a folder-form skill (it then moves to the Skills manager), and Delete.

Commands vs. skills

Flat commands/<name>.md files are the legacy single-file format; folder-form skills with a SKILL.md are the richer replacement (frontmatter, supporting files, invocation modes). The two widgets split along exactly that line, and "Upgrade to skill folder" is the migration path.

Agents manager

No default shortcut — open it from the command palette (Ctrl+Shift+P / Cmd+Shift+P, search "agents") or the quick-actions menu. It manages agent definition files: flat <name>.md files in .claude/agents/ (project) and ~/.claude/agents/ (personal).

Cards show the agent's description plus chips for its tool list and model (both from frontmatter), with the same shadowing marker as skills. The inline editor, creation dialog, duplicate/delete context-menu actions all work like the Skills manager; creation templates are Blank, Researcher (read-only research), Reviewer (code review with severity-grouped findings), and Implementer (makes a change and verifies it).

Two agent-specific extras:

• Eye toggle — each card has an eye button that hides the agent from the # picker's suggestions (and shows it again). Handy for agents Claude should be able to use but you never invoke by hand.
• Invocation pattern — a collapsible settings section at the top sets the phrase inserted when you pick an agent from #. Default: Consult with agent {agent}, with {agent} as the name placeholder.

Not the Sub-agents widget

The separate Sub-agents widget is a runtime monitor of Task-tool sub-agents spawned during the current session — see Cost analytics & history. This widget manages the definition files those agents are built from.

Plugins manager

No default shortcut — open it from the Skills manager's header, the command palette, or quick actions. It browses the Claude Code plugin marketplace and manages what's installed, shelling out to the claude plugins CLI under the hood.

Filter pills split the list into All / Installed / Available / Enabled / Disabled. Each card shows the plugin's status badge, description (click to un-clamp), source marketplace, install count, author, and a breakdown of what it ships — skills, agents, commands, hooks, MCP servers. The buttons on the card do the work: Install for available plugins; Enable/Disable and Uninstall for installed ones.

Plugins are the install unit; their components are not. Once a plugin is installed, its skills, agents, and commands appear automatically in the other managers (marked with a plugin badge, read-only).

Snippets

No default shortcut — open it from the command palette or quick actions, or via the Edit link in the # picker itself. Snippets are reusable text fragments inserted by typing # in chat; they are stored server-side, so they follow you across browsers and devices.

Each row shows the snippet's name, its #name trigger, and a preview of the text, with four actions: insert into chat, send now (inserts and sends in one click), edit, and delete — plus the same eye toggle as agents to hide a snippet from # suggestions without deleting it. Drag the handle on the left to reorder; the # picker follows your order. Click anywhere else on a row to edit it.

A snippet is just a name and an optional text body — if you leave the text empty, the name itself is what gets inserted.