Skills, tools & integrations pages

Three dashboard routes in the Operations section answer “what can my agent do, and how do I extend it?” The Skills page (/skills) is the only one you edit — create skill bundles by hand, or install them from the ClawHub marketplace and from GitHub. The Tools page (/tools) and Integrations page (/integrations) are read-only references: Tools lists every callable tool and CLI binary the runtime discovered, and Integrations is the status catalogue of channels and connectors.

Reach for this page when you want to add a capability (Skills), check whether a tool or binary is actually available to the agent (Tools), or see which integrations are wired up versus available to configure (Integrations). All three live behind the dashboard’s bearer-token auth — see Run the dashboard to pair a browser first.

Skills page (/skills)

A skill is a reusable instruction/capability bundle — Markdown instructions plus optional tool definitions — stored as an item in your Kumiho memory and on disk under ~/.revka/workspace/skills/. The Skills page is the registry manager for those bundles, with two extra panels for pulling skills in from outside: the ClawHub marketplace and a GitHub URL installer.

The page is a three-column layout:

• Left — a registry-status strip (total / active / deprecated / domain count) and a searchable index of your local skills. Search matches name, description, domain, and tags.
• Centre — the selected skill’s workspace: domain, revision number, created date, tags, description, and the skill content rendered as Markdown. From here you Edit, Deprecate/Re-enable, or Delete.
• Right — the GitHub installer and the ClawHub marketplace panel.

Manage local skills

Skills are CRUD entities backed by the gateway’s skills API. Every write creates a new revision.

Action	Method + path	Auth
List skills	GET /api/skills?include_deprecated=true&page=1&per_page=100	Bearer
Skill detail (with content)	GET /api/skills/{kref}	Bearer
Create skill	POST /api/skills	Bearer
Update skill (new revision)	PUT /api/skills/{kref}	Bearer
Deprecate / re-enable	POST /api/skills/deprecate	Bearer
Delete	DELETE /api/skills/{kref}	Bearer

The create/edit modal has five fields. name, description, and content are required; domain defaults to Other; tags is a comma-separated list that is split and trimmed on save.

Field	Type	Meaning
name	string	Skill name (required)
description	string	Short summary shown in the index (required)
content	string	Markdown body — the skill’s instructions/prompt (required)
domain	enum	One of Memory, Creative, Privacy, Graph, Behavioral, Other
tags	string[]	Comma-separated keywords for search

Creating a skill posts the form straight to the registry:

POST /api/skills
Authorization: Bearer rk_<token>
Content-Type: application/json

{
  "name": "release-notes",
  "description": "Draft release notes from a changelog",
  "content": "# Release notes\nSummarise the changelog into...",
  "domain": "Creative",
  "tags": ["writing", "release"]
}

The gateway stores the content as a SKILL.md artifact pointing at ~/.revka/workspace/skills/<slug>.md, with the description, domain, and tags held in the revision metadata. The kref used in the URL paths is the part after kref:// — for example CognitiveMemory/Skills/my-skill.skill.

Note

The content field is rendered as Markdown (GitHub-flavoured) in the viewer, so you can use headings, lists, and code fences. Deprecating a skill hides it from the agent without deleting it; Delete is permanent. The registry is cached server-side for ~30 seconds and the cache is invalidated on every write, so your changes appear immediately.

ClawHub panel

ClawHub is the community skill marketplace. The right-rail panel loads trending skills on first open; if trending returns nothing it falls back to a search for agent. Type a query and press Enter (or click Search) to search the marketplace — searches fire once the query is longer than one character. Click Trending to clear the query and reset to the trending list. Each result card shows a name and description with an Install button.

Action	Method + path	Auth
Search marketplace	GET /api/clawhub/search?q=<query>&limit=12	Bearer
Trending skills	GET /api/clawhub/trending?limit=12	Bearer
Skill detail	GET /api/clawhub/skills/{slug}	Bearer
Install a skill	POST /api/clawhub/install/{slug}	Bearer

Installing fetches the skill’s SKILL.md, writes it to ~/.revka/workspace/skills/<slug>.md, registers it as a skill item in your Kumiho memory project’s Skills space (tagging the revision published), and refreshes the local index so the new skill appears on the left immediately.

POST /api/clawhub/install/rust-debugging
Authorization: Bearer rk_<token>

{ "installed": true, "name": "rust-debugging", "slug": "rust-debugging" }

Caution

If the ClawHub integration is disabled the routes return 400 with {"error":"ClawHub integration disabled"}; an unreachable marketplace returns 502. Anonymous browsing and installing work without a token — a clh_… API token is needed only for publishing. Configure the integration under [clawhub] in config.toml (enabled, api_url, optional api_token); see Cost, audit, ClawHub & credentials API.

GitHub installer

The GitHub panel installs a skill directly from a repository URL. It reuses the same install endpoint — the pasted URL is passed where a slug would go:

POST /api/clawhub/install/{github-url}
Authorization: Bearer rk_<token>

1. Paste a GitHub URL into the installer field.
2. Press Enter or click Install.
3. On success the skill is registered into your local Kumiho Skills space and the index refreshes — exactly like a ClawHub install.

For the skill file formats (SKILL.md front-matter and SKILL.toml), the on-disk layout, shell/HTTP tool kinds, and the open_skills community library, see the full Skills system reference and revka skills, workflows & sop.

Tools page (/tools)

The Tools page is a read-only reference of everything the agent can invoke right now. Tools are registered at runtime — from built-ins, configured integrations, connected MCP servers, and discovered binaries — so this page reflects the live registry, not something you edit here. A single search box filters both sections at once, and a Reload button re-fetches them. The page has two collapsible sections.

It pulls from two endpoints:

Section	Method + path	Auth	Shows
Agent tools	GET /api/tools	Bearer	Tool name, description, and input-schema parameters
CLI tools	GET /api/cli-tools	Bearer	Binary name, path, version, category

Agent tools vs CLI tools

These two lists answer different questions, and the distinction matters when you are debugging “why can’t the agent do X?”

• Agent tools are the structured tools the model calls inside the agent loop — the MCP-style tool surface. Each is a card with a name and description; click a card to expand its full JSON parameter schema. This is the set the LLM sees and can call directly, including built-ins (shell, file_read, web_search_tool, …), credential-gated integration tools (notion, jira, google_workspace, …), and tools proxied in from external MCP servers (prefixed <server>__<tool>).
• CLI tools are external binaries on the host that the agent can shell out to (via the shell tool). The table lists each binary’s name, filesystem path, detected version, and a category tag. A binary appearing here means the runtime found it on PATH; it does not add a new model-callable tool — the agent reaches it through shell.

In short: agent tools are things the model invokes by name with typed arguments; CLI tools are programs on disk the model can run as shell commands. If a capability you expect is missing from the agent-tools list, it usually means the integration’s credentials aren’t set or its MCP server isn’t connected — not something fixable from this page.

Note

The agent-tool registry has baseline tools (always on, no credentials), extended tools (registered when matching credentials exist), and runtime-aware tools (cron, delegate, SOP, sessions, …) that need live runtime handles. For the full registry, deferred loading, and the tool_search mechanism, see Connecting to external MCP servers and Tools overview.

Integrations page (/integrations)

The Integrations page is the read-only catalogue of every integration Revka knows about — chat channels, AI-model providers, productivity connectors, and more — each with a status badge. It is informational: it tells you what is wired up and what is available to configure, but you do not activate anything here.

It is backed by one endpoint:

GET /api/integrations
Authorization: Bearer rk_<token>

The layout has a left rail (coverage counts + category filter), a centre grid of integration cards, and a right detail panel for the selected card. Click a category to filter; click a card to inspect its description, category, status, and operational readiness. Categories are derived from the catalogue itself and include Chat, AI Model, Productivity, Music & Audio, Smart Home, Tools & Automation, Media & Creative, Social, and Platform.

Status badges

Every integration carries one of three statuses:

Status	Badge	Meaning
Active	Active	Configured and running now (e.g. always-on Shell, File System, Weather, or a channel you’ve enabled)
Available	Available	Supported and ready to configure — not yet enabled
ComingSoon	Coming soon	Catalogued but not yet usable

From Available to Active

This page does not enable integrations — activation happens in configuration. To move an integration from Available to Active:

1. Note the integration and its category on this page.
2. Open the Config page (/config) and add or enable the matching section in config.toml — for example a [channels_config] entry for a chat channel, a [notion]/[jira]/[google_workspace] block for a productivity connector, or a provider block for an AI model.
3. Provide the required credentials (config key or environment variable), then save.
4. Reload the Integrations page — the status flips to Active once the runtime registers it.

Note

Channel configuration is covered in Connect a messaging channel and the Channels overview; productivity connectors (Notion, Jira, Composio, Google Workspace, Microsoft 365, LinkedIn) and the full status catalogue are detailed in the Integration catalog and Config: channels, tools & integrations.

Related pages

Skills system SKILL.md / SKILL.toml formats, shell & HTTP tool kinds, and the open_skills library.

Integration catalog The full status catalogue by category and what each integration needs.

Tools overview The complete agent tool registry — built-ins, integrations, and MCP tools.

Cron, cost & config pages Edit config.toml and activate channels and connectors from the dashboard.