We shipped an agent that reads every call on your pipeline and tells you which five deals to focus on Monday

Open your Attio pipeline. Count the deals that are still open. Now answer one question: which five are about to slip?

Most teams cannot. Not because the answer is hard, but because it is buried. It is in a call recording from nine days ago where the champion stopped saying "we" and started saying "they." It is in an email thread that went quiet. It is in a note nobody re-read.

We built an Attio deal health agent that reads all of it. We call it Deal Focus. It reads every call, email, and note on every open deal in your workspace, scores each deal red, yellow, or green, writes down why, and sends you the five that need you this week. Every Monday, before your pipeline review.

This post is what it watches for, how it scores deal health, how it runs, and how to put it on your own workspace.

Why your forecast is a guess

Attio has a native deal health score. It is a useful field. It counts recent email activity, upcoming meetings, and time in stage, and turns that into a number.

It is also blind to the thing that actually kills deals.

A deal can have fresh email activity and a meeting on the calendar and still be dead. The emails are the champion forwarding your message to a procurement team that will never reply. The meeting is a courtesy call to let you down easy. The activity count went up. The deal went to zero.

The signal that a deal is dying is not in the activity count. It is in what was said. "We need to loop in legal." "Honestly, the other option came in lower." "Let me check if the budget is still there." A person who read the last call would catch all three. A formula that counts emails catches none of them.

Reading the calls is the job. Nobody has time to do it for every deal, every week. That is the agent.

What the agent watches for

For every open deal above the threshold you set, the agent reads the call recordings, the email threads, and the notes, and looks for the signals that a deal is slipping.

• Unresolved objections. A concern was raised on a call and never closed. It is still sitting there.
• A decision-maker going quiet. The person who has to sign stopped showing up to calls or stopped replying.
• Competitor mentions. A rival came up, and how it came up. "We looked at them and passed" is green. "They came in lower" is red.
• Timeline slippage. The close date the deal was sold on does not match what was said on the last call.
• Champion language. The shift from "we are doing this" to "they are still deciding." It is the earliest signal there is, and it never shows up in an activity count.

How it scores deal health

Every open deal gets one of three colors.

Green means the conversations support the close date. Yellow means something is unresolved and a human should look. Red means the agent found a specific reason the deal is at risk.

The color is not the point. The reasoning is. The agent writes a short note onto the deal record in Attio explaining the score: which call, which sentence, which signal. You are not asked to trust a number. You are handed the evidence, and you decide.

Because the score and the reasoning live on the record, the whole team sees the same picture. The rep, the manager, and anyone who opens the deal a month later all read the same call-grounded summary instead of guessing from the activity feed.

The Monday digest

Every Monday morning the agent posts the five deals that need attention this week into Slack.

It is not a list of every red deal. It is a ranked, five-item focus list, and each item comes with a specific next action. Not "follow up." Something closer to: "The security review raised three calls ago is still open. Send the SOC 2 report and get their IT lead on a call before Thursday, or the close date does not hold."

The digest lands before your pipeline review, not after. The review stops being a status read-out and starts being a decision meeting, because the reading already happened.

How it runs

The Deal Focus agent ships as a Claude Code skill. The customer installs it once, schedules it once, and never touches the infrastructure again.

The runtime stack is short:

• color:var(--color-text-heading)]">The skill is a markdown file in a public GitHub repo at [github.com/gmaramigin/attio-deal-focus-skill. No Python runner. No Anthropic API key. No server to host.
• The reader is the Attio MCP server, already connected to your Claude Code. The agent uses it to search and read call recordings, emails, and notes, and to write the health score and reasoning back to each deal.
• The schedule is /schedule inside Claude Code. The run happens as a Claude Code routine on Anthropic's cloud, so the agent fires every Monday whether or not anyone's laptop is open.

The customer's Claude Code subscription is the entire runtime. We do not host the agent. We do not hold an API key. We do not see your pipeline. If we disappear tomorrow, the agent keeps running.

This is the rule for every agent we ship. Pipeline Hygiene works this way. Content Mining works this way. The customer owns the runtime, the data, and the schedule.

Why an agent beats a dashboard

You can already build a deal health dashboard in Attio. Filter on stage, sort by last activity, color by the native score. Many teams have one.

The dashboard has a problem. It shows you everything. Forty deals, all colored, all asking for a look. The work of deciding which five matter is still yours, and it is still the work nobody has time for on a Monday morning.

The agent does the deciding. It reads, it ranks, it hands you five, and it tells you the first move on each. A dashboard makes the data visible. An agent makes the call. That is the difference between knowing your pipeline is messy and knowing what to do about it before lunch.

What it costs

Build. A one-time build inside your workspace. The skill installed on your Claude Code, the risk signals tuned to how your team sells, the threshold set, the Slack channel wired, and the schedule live.

Run. An optional monthly retainer if you want us watching the output, tuning the signals, and adjusting the scoring as your sales motion changes.

Claude Code consumption. The agent runs on your existing Claude Code subscription. No separate Anthropic API key. No usage-based bill from us.

Pricing for both options is on the pricing page.

How to install

Three steps if you want to install it yourself.

1. Clone the skill repo into your Claude Code skills directory. git clone https://github.com/gmaramigin/attio-deal-focus-skill ~/.claude/skills/attio-deal-focus-skill.
2. Connect the MCPs for Attio and Slack on your Claude Code. Each takes about five minutes the first time.
3. Schedule the run. Inside Claude Code: /schedule "Run the attio-deal-focus-skill in scheduled mode" --cron "0 8 * * 1". That is 8 AM Monday in your local timezone.

If you want it live this week without doing the install yourself, book a call and we will set it up on your workspace.

Start here if your forecast is a feeling

Every sales team says it has a forecast. Most have a feeling, sorted into stages.

The fix is not more discipline about updating stages. It is reading the conversations, every week, on every deal. That has always been too much work for a person to do by hand. It is the right amount of work for an agent.

If you run Attio and your pipeline review starts with "so, where are we," the Deal Focus agent is the one to ship first.