Skills

A Skill is a reusable system prompt — optionally paired with a whitelist of MCP tools — that you trigger by typing /<skill-name> in the Griffin AI Search input. When you invoke a skill, the LLM call for that turn is prefixed with the skill's system_prompt, and if the skill has a tool whitelist set, the tool registry for that turn is restricted to only those tools. Skills are a way to package a prompt you'd otherwise retype (and tool boundaries you'd otherwise have to remember to set) into something you can pull in with a few keystrokes.

A typical reason to create one: you always want CVE triage done the same way — the same criteria, the same output shape — so you write that once as a skill's system prompt, name it something like triage-cve, and from then on type /triage-cve instead of re-explaining your criteria in every conversation. If you also want to guarantee the model can't reach for anything outside a specific set of tools while triaging, you attach an MCP tool whitelist to the same skill so that turn's tool registry is scoped down accordingly.

What a Skill is made of

Example

A CVE-triage skill might be filled in like this:

• Name: Vulnerability Triage
• System prompt: You are a vulnerability triage assistant. Respond in formal regulatory tone, cite frameworks.
• Allowed MCP tools: safeguard_find_vulnerabilities, safeguard_list_projects
• Public: off, if this is just for your own workflow — on, if the whole team should triage the same way.

With this saved, typing / in Griffin AI Search and selecting Vulnerability Triage from the picker inserts /Vulnerability Triage as a token in your message and attaches that system prompt to the turn — restricting the tool registry to just those two tools, no matter what else is available.

Creating and managing Skills

Skills are managed from the Skills tab on the Integrations page:

1. Go to Integrations → Skills.
2. Click New skill to open the create dialog, and fill in a name, an optional description, a required system prompt, and (optionally) an MCP tool whitelist and whether the skill should be public.
3. Save. The skill now appears in your Skills table and is available via /<name> in Griffin AI Search.

The Skills table lists the skills you've created. Clicking a row (or the eye icon) opens the skill's detail panel inline, where you can view and edit every field, including the system prompt and tool whitelist. The detail panel has two tabs:

• Overview — a read-only summary: visibility (Public/Private), how many tools are whitelisted (or "Unrestricted" if none), created/updated timestamps, the full system prompt, the tool whitelist as a list of badges, and which conversations have used this skill.
• Edit — the same fields as the create dialog (name, description, system prompt, allowed MCP tools, public toggle), saved with Save changes.

From the detail panel you can also click New conversation to start a fresh chat in Griffin AI Search with this skill already attached — useful when you want to jump straight into using a skill without typing /<name> yourself.

From the table you can also:

• Toggle Enabled/Disabled directly on a row — you'll see a confirmation toast ("Skill enabled" / "Skill disabled") and the change takes effect immediately.
• Delete a skill via the trash icon — this asks for confirmation (naming the skill, e.g. "Delete skill "Vulnerability Triage"? This cannot be undone.") before removing it permanently.

If you haven't created any skills yet, the table shows an empty state prompting you to create one to bundle a reusable system prompt.

Public skills created by other users in your tenant don't show up in this table — that list is scoped to skills you created yourself. Shared/public skills from others are surfaced separately (see below and in the / picker).

Note that the Allowed MCP tools field is currently a plain text list, not a picker against your connected MCP tools — you need to type each tool's exact name (comma-separated), matching what's registered on the MCP Connectors page. A typo or near-miss name is simply excluded from the whitelist rather than flagged as an error.

Invoking a Skill with "/"

While typing in the Griffin AI Search input, type / to open the skill picker. It lists the skills available to you — both your own and any public skills shared tenant-wide — searchable by name, with each entry showing its name and description. A skill created by someone else and shared as public is labeled Public in the list; your own private skills show no such label.

From the picker you can:

• Type to filter the list down by name.
• Select a skill to attach it to your current turn — the model's response for that turn is generated using that skill's system prompt (and, if the skill has one, its restricted MCP tool set).
• Choose Create new skill if nothing existing fits, which opens the same create dialog described above without leaving the search box.

If you haven't created or been given access to any skills yet, the picker shows an empty state rather than a list.

Public vs. private skills

Skills are user-scoped by default — a skill you create is visible only to you, in your own Skills table and your own / picker.

Setting a skill's Public flag shares it tenant-wide: everyone else in your tenant then sees it (marked Public) in their / picker, so they can invoke it the same way you do. Making a skill public shares the prompt and tool whitelist for others to use — it doesn't move ownership of the skill itself, which is why public skills created by others appear as a distinct set from the ones you manage directly in your Skills table.

Use a private skill for something personal to your own workflow, and a public skill for something you want your whole team invoking consistently — like a shared CVE-triage prompt everyone should be running the same way.

FAQ & Troubleshooting

I invoked a skill but the model still used a tool I expected it to avoid. Check the skill's Allowed MCP tools list in its detail panel. If that list is empty, no tool restriction is applied for that skill — the turn's tool registry is unrestricted. Add the specific tools you want available to narrow it down.

I disabled a skill but can still see it in the picker. A disabled skill is skipped at compose time, not hidden from the UI — you can still see and re-enable it, but invoking it while disabled has no effect on the model's prompt or tool access. Re-enable it from the Skills table if you meant to keep using it.

A teammate says they can't see a skill I created. Check whether the skill's Public flag is set. Skills are private by default and only visible to their creator; mark the skill public to share it tenant-wide.

I deleted a skill by mistake. Deletion is permanent and confirmed via a dialog naming the skill before it proceeds — there's no undo, so recreate it with the same name, description, system prompt, and tool whitelist if you need it back.

Why don't I see other people's private skills in my Skills tab? The Skills tab only lists skills you created. Public skills shared by others are surfaced separately — through the / picker in search, where they're labeled Public — rather than mixed into your own management table.

I added a tool to the whitelist but the model still can't use it. Since the field is free text (see above), a misspelled or slightly-off name is silently dropped rather than flagged. Open the skill's Edit tab and re-check each entry against the exact tool names on the MCP Connectors page.

Related

• Griffin AI — the underlying model that Skills prompt and Griffin AI Search runs on.
• Search — the Griffin AI Search interface where / invokes a skill.
• Integrations — the Settings area, including the Skills tab, where connected systems and skills are managed.
• MCP Connectors — the MCP tools a skill's allowlist can restrict a turn to.