Settings: Capabilities

The Capabilities panel (/settings/capabilities) merges two formerly separate panels into a single tabbed page: Skills (the old Skills Hub, for managing skill packs) and MCP & Voice (the old Tools page, for MCP servers, image generation, and speech-to-text).

The left sidebar entry is labelled Skills & Tools and navigates to /settings/skills (the standalone Skills page). This Capabilities panel is at /settings/capabilities and is reached via the router directly or via the /settings/tools redirect. See the URL table below for the full redirect map.

Note

MCP server management was moved out of the modal in a previous release. The MCP & Voice tab now contains only a link to the full MCP Library plus the image generation and speech-to-text sections. For browsing, installing, and configuring individual MCP servers, see Settings: MCP Library.

Skills tab

The Skills tab embeds SkillsHubSettings without its outer wrapper. This is the older skill hub component; the standalone /settings/skills route uses a newer SkillsSettings component with a different layout (filter rail, health bar, detail drawer). The controls available in this tab are described below.

Discovered external skills

When Wayland finds skills in the directories of other CLI tools, they appear in this section before your own library. Skills are scanned from the following locations:

Source	Path
Global Agents	~/.agents/skills
Gemini CLI	~/.gemini/skills
Claude Code	~/.claude/skills
OpenCode	~/.config/opencode/skills
OpenCode (alt)	~/.opencode/skills
Custom paths	Paths you register via “Add Custom Skill Path”

Custom paths are stored in custom_external_skill_paths.json inside the app’s working directory. Each path is validated against an allowlist of approved directories at both registration time and enumeration time; paths outside that allowlist are silently skipped.

Import a discovered skill to copy it into your user skills directory. Import All copies every skill from the active source tab in one step. Refresh rescans the above locations without restarting.

My skills

Your installed skills. Each row shows the skill name, description, and a badge: Built-in (shipped with the app) or Custom (added by you). Hover a row to reveal Export and Delete actions.

• Export copies the skill (as a symlink) to a discovered external source directory you choose from a dropdown. Available only when at least one external source is detected.
• Delete is available only for custom skills. Built-in skills cannot be removed here.

The path to your user skills directory is shown below the toolbar:

<app-data>/skills/

Built-in skills are kept in:

<app-data>/builtin-skills/

Both paths are resolved at runtime by the app and displayed in the UI for reference.

Extension skills

Skills contributed by installed extensions appear in a separate section, badged to distinguish them from built-in and custom skills. They have no delete action here; uninstalling the extension removes them.

Auto-injected skills

Skills listed here are injected into every agent automatically. They are app-managed and cannot be toggled or removed from this panel.

MCP & Voice tab

The MCP & Voice tab has three sections: MCP Servers (a redirect to the MCP Library), Image Generation, and Speech-to-Text.

MCP Servers

This section links out to the MCP Library full-page view (/settings/mcp-library/installed). The old inline CRUD for adding and editing server rows was removed when the MCP Library page was introduced. Configure MCP servers there.

Image generation

Enable or disable the built-in image generation MCP server. The toggle requires a model to be selected first; it is disabled when no image-capable model is configured.

Control	Description
Model selector	Choose a provider and model. Only providers that include a model whose name contains image, banana, or imagine appear in this list.
Enable toggle	Starts or stops the built-in image generation server and syncs it to all agents.
Agent status	Badge showing which agents the server is installed for, visible when the server is enabled.

The selected provider, base URL, API key, and model name are written to the built-in server’s transport.env as WAYLAND_IMG_PLATFORM, WAYLAND_IMG_BASE_URL, WAYLAND_IMG_API_KEY, and WAYLAND_IMG_MODEL. The API key is kept in sync with the key stored on the provider record in Models settings; if you remove a provider, the image generation config clears automatically.

Speech-to-text

Enable voice input for the chat box. Configuration is stored at tools.speechToText in app config.

Provider	Required fields	Notes
OpenAI	API key (via Providers)	Uses whisper-1 by default. Base URL and language are optional. The API key is entered in Models settings, not here; a “Open Providers” button links there.
Deepgram	API key	Uses nova-2 by default. Base URL, model, language, detect language, punctuate, and smart format are configurable.
Whisper (local)	None	Runs on-device. Choose base or small model. The model file downloads from Hugging Face (ggerganov/whisper.cpp) on demand. The UI shows an installed state with a re-download option.

A microphone check widget is always shown after selecting a provider, so you can confirm your input device is working before relying on voice input.

Caution

The Whisper local download shows a progress bar animation but does not yet report real download progress. The label reads “Downloading… (progress reporting coming soon)”.

Whisper model locations

Model	Asset ID	Source URL
base	whisper-ggml-base	https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin
small	whisper-ggml-small	https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.bin

Install state is checked via voiceAsset.exists on load and after any download. A cancel button stops an in-progress download.

Text-to-speech

Enable voice output. Configuration is stored at tools.textToSpeech in app config.

Control	Description
Provider	kokoro-local (on-device) or system-native (OS speech synthesis).
Voice	Free-text voice name passed to the provider.
Speed	Slider from 0.5x to 2x in 0.1 increments.
Auto-read responses	When enabled, assistant responses are spoken automatically.
Test voice	Plays a short phrase via window.speechSynthesis regardless of stored provider, as a device sanity check.

When kokoro-local is selected, a Download Model button fetches the Kokoro ONNX model from:

https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/kokoro-v1.0.onnx

Asset ID: kokoro-onnx-model. Install state is probed on mount and clears to a re-download option once the file is present.

Caution

Like Whisper, the Kokoro download does not yet report real progress.

URL and tab switching

URL	Result
/settings/capabilities	Opens on the Skills tab (default).
/settings/capabilities?tab=skills	Opens the Skills tab.
/settings/capabilities?tab=tools	Opens the MCP & Voice tab.
/settings/skills-hub	Router redirect to /settings/skills (the full Skills page, not this panel).
/settings/tools	Router redirect to /settings/capabilities?tab=tools.

Tab switches update the URL in place (replace, not push) and preserve other query parameters.

The settings modal sider uses a separate remap table (LEGACY_ANCHOR_REMAP in SettingsSider.tsx). In that context, the tools and mcp anchor IDs both resolve to the mcp-library sider entry, and the capabilities anchor resolves to skills. These remap only when navigating within the modal; the router paths above apply to full-page settings navigation.

Related

• Settings: MCP Library
• Settings: Skills
• Settings: Voice
• Settings: Images
• Settings: Models and Providers