Skill Discovery

Agents can discover, browse, and request skills from the full catalog — not just the ones already assigned to them. Combined with connector-based suggestions, the platform proactively surfaces relevant skills as your team evolves.

How It Works

MeetLoyd uses three channels to surface relevant skills for your agents:

┌──────────────────┐   ┌──────────────────┐   ┌──────────────────┐
│  Agent Self-Serve │   │  First-Start     │   │  Continuous       │
│  (MCP tools)      │   │  (onboarding)    │   │  (watchdog)       │
└────────┬─────────┘   └────────┬─────────┘   └────────┬─────────┘
         │                      │                       │
         ▼                      ▼                       ▼
  Agent browses          Auto-assign during      Periodic check for
  catalog & requests     team setup               new matches

1. Agent self-service — Agents use skill_discover to browse the catalog and skill_request to ask for assignment
2. First-start suggestions — During team onboarding, the platform matches skills to each agent's tools and role
3. Watchdog monitoring — Periodic checks detect when new skills become relevant (e.g., after catalog updates)

All skill assignments go through human approval, except for first-party MeetLoyd skills during onboarding which are auto-assigned.

Browsing the Skill Catalog

Agents can browse the full skill catalog using the skill_discover tool:

// Browse all skills
{ name: "skill_discover" }

// Filter by category
{ name: "skill_discover", arguments: { category: "engineering" } }

// Search by keyword
{ name: "skill_discover", arguments: { query: "sales objection" } }

The response includes lightweight metadata for every matching skill:

{
  "skills": [
    {
      "id": "coding-agent",
      "name": "coding-agent",
      "description": "Full-stack software development capabilities...",
      "category": "engineering",
      "author": "meetloyd",
      "assignedToMe": false
    },
    {
      "id": "terraform",
      "name": "terraform",
      "description": "Terraform/OpenTofu IaC...",
      "category": "engineering",
      "author": "meetloyd",
      "assignedToMe": true
    }
  ]
}

Each skill shows assignedToMe so the agent knows what it already has. Only name and description are returned — no instructions are loaded until the skill is assigned and queried.

Requesting a Skill

When an agent finds a useful skill, it can request assignment:

{
  name: "skill_request",
  arguments: {
    skillId: "coding-agent",
    reason: "I need file editing capabilities to complete the refactoring task"
  }
}

This creates a task for an administrator to review and approve. The agent receives a task ID and can check status later.

important

Agents cannot load skill instructions until assignment is approved. Browsing shows metadata only.

Connector-Based Suggestions

When integrations are connected, the platform automatically suggests relevant skills.

How Connectors Map to Skills

Integration	Suggested Skills
GitHub	coding-agent, webapp-testing
Jira	webapp-testing
Slack	internal-comms
Google Workspace	doc-coauthoring, xlsx
Microsoft 365	doc-coauthoring, docx, pptx, xlsx
Notion	doc-coauthoring
HubSpot	internal-comms

Automatic Suggestions on Connect

When you connect a new integration, the platform:

1. Looks up recommended skills for that integration
2. Checks which skills exist in your catalog
3. Creates a review task with suggestions
4. An admin reviews and assigns to relevant agents

This ensures agents always have the right skills for the tools available to them.

First-Start Discovery

During team onboarding (Team Starting Procedure), skill discovery runs automatically after tool discovery:

1. Tool Discovery — Identifies which integrations are connected
2. Skill Discovery — Matches agents to skills based on:
   • Their assigned tools (e.g., hubspot_ tools suggest sales skills)
   • Their role description (keyword matching)
   • Connected integrations (connector-to-skill mapping)
3. Auto-Assignment — First-party MeetLoyd skills are assigned automatically
4. Suggestions — Third-party skills create review tasks for the admin

Managing Suggestions

In the Dashboard

Skill suggestions appear in your team's Activity Feed and as tasks in the Tasks panel. To review:

1. Open Tasks in the dashboard
2. Look for tasks titled "Review skill suggestion: ..."
3. Approve to assign the skill, or dismiss

Via API

// List pending skill suggestion tasks
const tasks = await client.tasks.list({
  type: 'manual',
  input: { requestType: 'skill_suggestion' }
});

// Approve by assigning the skill
await client.skills.assign({
  skillId: 'coding-agent',
  agentId: 'agent-123'
});

Discovering Integrations

Agents can also discover what integrations are available using connector_discover:

// See all connected and available integrations
{ name: "connector_discover" }

// Filter by category
{ name: "connector_discover", arguments: { category: "developer" } }

And request missing integrations:

{
  name: "connector_request",
  arguments: {
    integrationId: "github",
    reason: "Need GitHub access for code review tasks"
  }
}

Best Practices

For Admins

• Review suggestions promptly — Agents work better with the right skills
• Trust auto-assignment — First-party skills are tested and safe to auto-assign
• Check the activity feed — Skill discovery events show what was matched and suggested

For Skill Authors

• Use clear categories — Skills with accurate categories get better auto-discovery matches
• Write descriptive names — Keyword matching works on skill names and descriptions
• Tag with metadata — Tags help the scoring algorithm surface your skill to the right agents

---

Next: Learn about the Coding Agent for giving agents file editing and bash capabilities.