How Skill Routing Works

Every time you send a message in Agency Hero, something happens in the background before the agent responds: the platform selects the most appropriate skill to handle your request. This process is called skill routing, and it’s what allows one chat interface to intelligently handle everything from meeting prep to deal research to writing documentation.

This article explains how routing works, how workspace settings shape what’s available, and when you might want to step in and choose a skill yourself.

What is a skill?#Copy link

A skill is a focused AI module that defines:

• The role the agent plays (e.g., researcher, task manager, document author)
• Which tools it can access (e.g., web search, Linear, meeting notes)
• What guardrails apply to keep it on-task and within permitted boundaries

Think of skills like specialist team members. Each one is optimized for a specific type of work. Skill routing is how Agency Hero decides which specialist picks up your message.

Related: Understanding AI Agents

How routing works: the priority chain#Copy link

Every time you send a message, Agency Hero evaluates a short priority chain to determine which skill to load — before any tools are called or context is fetched. The checks happen in this order:

In practice, most everyday messages land at Priority 3 — the General skill handles them and delegates as needed.

The General skill: your default orchestrator#Copy link

The General skill is the hub of Agency Hero’s AI. It acts as an orchestrator — a generalist that can handle a broad range of tasks directly, and that knows when to hand off to a more specialized subagent.

When you ask something complex (“Do a deep dive on this prospect and draft a follow-up email”), the General skill coordinates the work by spinning up specialist subagents in the background. Each subagent has a tightly scoped toolset focused on one type of work:

Subagents always inherit the permissions of the conversation they were launched from — they can never access more than what you’re already allowed to do.

The General skill itself has a broad toolset (~38 tools), so for many straightforward requests it handles things directly without delegating.

Related: Creating and Managing Tasks via Chat

Forcing a specific skill with @skill-name#Copy link

Sometimes you already know which specialist you want. Instead of letting the orchestrator decide, you can invoke a skill directly by typing @ followed by the skill’s name in the chat composer.

Examples:

• @research-assistant Do a competitive analysis on Acme Corp
• @doc-author Rewrite this section in a more concise tone
• @meeting-prep Prep me for tomorrow's call with DataCo

How it works:

1. When your message begins with @, Agency Hero runs a fast lookup against all available skill slugs.
2. If it finds a match, that skill is loaded directly — bypassing the normal priority chain.
3. The skill is marked as an explicit selection, so the system knows you chose it intentionally.
4. If the @mention doesn’t match a valid skill slug, the message falls through to normal routing.

When is this useful? Use @skill-name when you want precision — for example, when you want a thorough independent research pass rather than having the orchestrator’s general response, or when you’re specifically writing a doc and want the Doc Author’s focused toolset from the start.

How your workspace shapes which skills are available#Copy link

Not every skill is available in every workspace. Skills are bundled into skill packs — groups of related skills that are assigned based on your workspace type.

A skill can only be routed to if it’s enabled in your current workspace. If you try @research-assistant in a workspace that doesn’t include it, the mention won’t resolve and your message will fall through to normal routing.

Workspace admins can customize which skills are active by going to Settings → Skills, where individual skill overrides can be added or removed.

If you expect a skill to be available but it isn’t responding to @mentions, reach out to your workspace admin — the skill may simply not be enabled for your workspace type.

What happens after a skill is selected: tool filtering#Copy link

Once a skill is chosen, there’s one more layer before the agent acts: tool filtering. Even within a skill’s allowed toolset, the exact tools available to the agent depend on three factors:

1. The skill’s own tool list — each skill defines exactly which tools it’s permitted to use.
2. Your workspace’s permission policy — some workspaces are configured in read-only mode or with blocked operations, which further restricts what tools can run.
3. Connected integrations — tools tied to external services are only active if that service is connected. For example, GitHub-related tools are removed if no repositories are connected; Linear task creation requires an active Linear integration with write access.

This layering means the agent is always working within boundaries appropriate to your context — no surprises.

Related: Tool Approvals and Permissions

Practical takeaways#Copy link

• Routing is automatic. For the vast majority of messages, you don’t need to think about it. The General skill handles your request and delegates to specialists as needed.
• Use @skill-name when you want control. If you know exactly which specialist fits your task, invoke it directly to skip the orchestrator’s decision.
• Skills vary by workspace. What’s available depends on your workspace type and any overrides your admin has configured.
• Tool access is layered. A skill’s tools are further filtered by workspace policy and connected integrations — what you can do depends on what’s been set up.
• Subagents can’t exceed your permissions. Any specialist the orchestrator delegates to inherits your permission level, never exceeds it.

Summary#Copy link

Skill routing is the process Agency Hero uses — automatically, on every turn — to match your message to the right AI module. Most of the time it’s invisible: the General skill picks up your request, delegates where needed, and returns a coherent result. When you want more precision, @skill-name puts you in the driver’s seat. And when something seems missing, the answer is usually a workspace configuration question your admin can help with.