Skip to main content

HubSpot Agent Library

Seven CRM maintenance agents for HubSpot Agent Hub

Full configuration for seven agents that keep a portal clean between cleanups: trigger, inputs, actions, instruction text, and what each one is never allowed to decide.

Download 2 free agents as Claude Skill (.zip)

Lifecycle Stage Auditor and Contact Sunset. No form needed.

The full set adds:

  • The other five agents, from duplicate triage to the monthly portal health report
  • All seven build sheets in one PDF guide, in build order
  • One Claude skill that covers the full set

Get all 7 agents here👇

What's inside

Seven agents (in the order we'd build them)

Company association fixer

Matches contacts that land without a company to the right one within a day.

Duplicate triage

A weekly queue of the duplicates HubSpot's own tool missed, with the evidence.

Lifecycle stage auditor

Catches contacts whose lifecycle stage contradicts what their deals say.

Owner coverage checker

Finds records owned by people who left and suggests who takes over.

Property normalizer

Maps free-text titles, countries and industries to your picklists as they arrive.

Contact sunset

Recommends which disengaged marketing contacts to stop paying for.

How to use it

Three steps

Start with one agent. Add the next once the first is earning its keep.

1. Download it

Grab the two free agents above, or fill in the form for the full set and the PDF guide.

2. Load it into your AI

Add the skill to Claude, or give the guide to the AI you already use along with your company context. It runs the readiness check with you and picks where to start.

3. Build one agent

Configure it in Agent Builder and test it on real records. Keep a person approving the output until eight in ten go through unedited.

Built to be careful

Three rules every agent follows

One job each

An agent that flags, scores and recommends is three agents. Narrow ones cost less and break where you can see it.

Limits in writing

Every set of instructions ends with what the agent must never decide.

Autonomy is earned

Every agent starts supervised. Some writes stay behind approval for good.

Before you build

Three questions decide whether an agent is worth its credits.

Every build sheet assumes these are true. If they aren't, the agent still runs, and you pay it to tell you what you already suspected.

Definitions

Do your lifecycle stages and picklists mean one thing?

An agent can only follow a rule that's written down. Undefined values get defined by the model, differently each run.

Activity

Is the work logged in HubSpot?

An agent can't tell never happened from never logged.

Ownership

Will one admin work the review queue every week?

If nobody reviews the proposals, you're paying for noise.

Two or more no's and the foundation comes first. The guide shows what to fix. If you'd rather not fix it alone, that's the work we do.