Skills

Skills are prompt templates with execution guardrails, delivered to agents through the MCP gateway as tool responses. Browse the skill library, install skills, or create custom ones for your organization.

Key Concepts

How skills work

Skills are prompt injections via MCP tools. When an LLM calls a skill tool (prefixed with skill__), hub-core loads the skill definition, renders the prompt template with the provided parameters, optionally injects project context, and returns the rendered prompt as the tool response. The LLM then acts on the prompt — no server-side orchestration is involved.

Skill portability

Because skills travel through the MCP gateway, they are portable across all LLMs. A skill installed for a Claude-powered agent works identically with a GPT or Gemini-powered agent. The skill definition is model-agnostic.

Prebuilt skills

Clevername ships with prebuilt skills for common tasks. These are seeded automatically and available to all agents. Browse the skill library to see what is available.

Step-by-Step Guide

Browse the skill library

Open Integrations in the dashboard sidebar and follow the Skills links, or go directly to /dashboard/skills. The skill library shows all available skills organized by category, including prebuilt skills and any custom skills created by your organization.

Skills library page showing skill cards organized by category with install buttons

Browse skills by category. Each card shows the skill name, description, and parameter count.

Install a skill

Click on a skill card to see its details — description, parameters, prompt template preview, and usage examples. Click Install to add the skill to your available tools. Installed skills become selectable in the agent's tool-scope configuration (questionnaire question T1).

Note

Installing a skill makes it available for assignment to agents. The skill is not active until you assign it to at least one agent through the agent's integrations configuration.

Create a custom skill

Click Create Skill (/dashboard/skills/create) to build your own. You'll need to provide:

Name — used as the tool name (automatically prefixed with skill__).
Description — shown to the LLM as the tool description.
Parameters — define the inputs the skill accepts (name, type, required).
Prompt template — the prompt text with parameter placeholders.
Category — for organizational purposes in the library.

Create Skill form showing fields for name, description, parameters, and a prompt template editor

Define parameters and write a prompt template. Use double-brace syntax for parameter placeholders.

Assign skills to agents

Once installed, skills are assigned through the agent's tool scope: select them in questionnaire question T1 when you submit for review, or from the agent's integration bindings on its detail page. Only the skills on the agent's allowlist can be invoked.

Important

If the Agent Review is enabled, adding or removing skills from an approved agent changes its integration bindings, which may trigger drift detection and require re-approval.

Monitor skill usage

Skill usage is tracked in the skill_usage_events table. Connected agents can read skill health through the company__read_skill_health gateway tool and record outcomes with company__log_skill_outcome.

AI Company Overview Creating Agents Deploying Agents Agent Review Overview