documentation
03 dashboard

Skills

Skills are packaged extensions that add capabilities to your agents. Browse the bundled set, install new ones from the marketplace or a local file, review them for safety, and control which agents are allowed to use them.

What this page is for

The Skills page is the management surface for everything that isn’t a built-in tool or an MCP server. If you’ve ever wanted to give an agent a capability that doesn’t ship in the default tool set, you either install an MCP server (fast, community-maintained) or you install a skill (more bespoke, runs inside Exolvra). This page is where skills live.

See Tools, skills & MCP for when to reach for a skill versus the other two.

Layout

The Skills page has two categories:

  1. Bundled skills — the ones that shipped with Exolvra. Always available, no approval needed.
  2. User-installed skills — skills you added yourself. Each has an approval status.

Each skill card shows:

  • Name and description
  • Version
  • Author
  • Status — Enabled, Disabled, Pending approval, or Rejected
  • Signed by — a badge if the skill is cryptographically signed
  • Used by — how many agents have this skill in their tool set

Click any skill to open its detail page with the full manifest, permissions, and action buttons.

Common tasks

Install a user skill from a file

  1. Click Install skill in the page header
  2. Pick a local .zip or directory containing the skill manifest
  3. Exolvra loads the skill and verifies its signature (if signed)
  4. The skill appears in the list with Pending approval status
  5. Review it — read the manifest, check which permissions it requests, verify the author if you know them
  6. Click Approve if you trust it. It becomes available for per-agent grants.

Never approve a skill you haven’t reviewed. Approving grants it access to whatever permissions its manifest declares — that could include filesystem, network, or shell.

Enable or disable a skill

Open the skill detail page and toggle Enabled. Disabled skills are loaded but inert — no agent can invoke them, even if they’re in that agent’s tool set. Use disable as a kill switch when you want to temporarily stop a skill without uninstalling.

Grant a skill to an agent

Per-agent grants live on the agent edit page, not here. Open the agent, go to the tools section, find the skill, check its box. From that moment the agent can call the skill’s capabilities.

You can also do the reverse: from the skill detail page, see the list of agents currently granted and revoke from the list.

Uninstall a skill

From the skill detail page, click Uninstall. Confirm. The skill is removed from disk and from the database. Any agents that had it granted lose access immediately on their next turn.

Bundled skills can’t be uninstalled — they come back on the next gateway restart.

Update a skill

If a skill author releases a new version, you can update by installing the new bundle. Exolvra detects that the id matches an existing skill and offers Update instead of Install new. Update is a version bump — all existing agent grants survive.

The approval flow

User-installed skills pass through an explicit approval:

  1. Installed — files on disk, manifest parsed, signature verified (if signed)
  2. Pending approval — visible on this page, not yet usable by any agent
  3. Approved — an admin reviewed it and clicked Approve. Now usable, subject to per-agent grants.
  4. Rejected — an admin reviewed it and declined. Marked as rejected; cannot be used. Can be reconsidered later.

Rejection is recoverable — you can change your mind and approve a previously rejected skill from the same page.

Admin-only

Only Admin and Owner roles can approve or reject skills. Regular users see the approval status but can’t change it. This is the main reason the approval flow exists — it keeps random users from installing arbitrary skills in a multi-user deployment.

Signed skills

Skills can be cryptographically signed. When a signed skill is installed, Exolvra verifies the signature against the declared public key and displays a Signed by badge on the skill card. Unsigned skills are still usable but don’t get the badge.

Treat signed + known-author skills as higher-trust than unsigned or unknown-author skills. The signature doesn’t make the skill safe — it just proves it came from the person it claims.

Common pitfalls

Approving every skill that gets installed. Approval is the safety gate. If you approve without reviewing, the gate doesn’t exist. Take 60 seconds to read the manifest before clicking approve.

Forgetting per-agent grants. Installing and approving a skill doesn’t automatically grant it to any agent — you still have to check the box on each agent that should have it. A just-approved skill is invisible to your agents until you wire it up.

Mixing up skills and MCP servers. Skills run inside the Exolvra process. MCP servers run as separate processes or over HTTP. Both extend agent capabilities, but they’re implemented differently. Check Tools, skills & MCP if you’re not sure which you need.

Where to go next