auto-skill-loader

Give the agent persistent, automatic knowledge of your approved skills at session start. No manual triggering needed.

How It Works

The MCP server exposes two things:

• `skills://active` resource — contains all your approved skills, read at session init
• Tools — for managing which skills are active

Tools

list_skills

Call list_skills() with no arguments. Returns all skills found in your skills directory with descriptions and active status.

Use when:

• User asks "what skills do I have installed"
• You need to see what's available before activating something

activate_skill

Call activate_skill(name="skill-name") to add a skill to the auto-load list. The skill must already be in your skills directory.

Use when:

• User wants to add a skill to their approved list
• A skill is installed but not auto-loading

deactivate_skill

Call deactivate_skill(name="skill-name") to remove a skill from the auto-load list.

Use when:

• User wants to stop a skill from auto-loading
• A skill is causing issues or conflicts

suggest_skills

Call suggest_skills() with no arguments. If no skills are active, this scans available skills and suggests common ones to get started.

Use when:

• User has no skills configured yet and needs help getting started
• First session on a fresh machine

On Session Start

At session start, the host (OpenCode/Claude Code/etc.) reads skills://active. This resource contains all skills currently in your config.yaml active list, concatenated together. The agent gets them injected automatically — no action needed from you.

If you make changes via activate_skill or deactivate_skill, they persist to config.yaml and will be active on next session.

No Explicit Triggering

Unlike skills that require skill({ name: "..." }) calls, auto-skill-loader works because:

1. The host reads skills://active at session init
2. Your approved skills are already in context before you say anything
3. The agent can reason about them without special invocation

Example Workflows

New user — first time setup

You: "what skills do I have?"
Agent: list_skills()
  → returns: vision-analysis, context-maintainer, markdown-mcp, etc.
You: "activate vision-analysis and context-maintainer"
Agent: activate_skill(name="vision-analysis")
  → ✅ Activated
  activate_skill(name="context-maintainer")
  → ✅ Activated
Agent: "Done! Both skills are now in your auto-load list. Restart the session to use them."

Check what's active

You: "which skills are auto-loading?"
Agent: get_active_skills()
  → returns list of currently active skill names

Remove a skill

You: "stop loading the markdown-mcp skill"
Agent: deactivate_skill(name="markdown-mcp")
  → ✅ Deactivated