<aside> 💡 Use case: someone messages your page late at night ready to buy, and by morning they have gone elsewhere. Or someone sends spam, and you lose sleep checking it. This playbook answers every message with words you approved in advance, logs real leads in HubSpot, and wakes you only when it matters.

</aside>

<aside> ⚡ What sets it off: a DM or lead arrives out of hours. You paste the message, or the out-of-hours inbox sweep passes it in. What counts as a real lead, and what is urgent, comes from the definitions you set in brand-os/conversation/routing-rules.md.

</aside>

What happens, step by step

<aside> 🧩 Department on this card: Leads & conversations.

  1. Orchestrator: checks the kill switch and the owner files, and writes one brief with the message and the time it arrived.
  2. Leads & conversations: sorts it: real lead, problem, question or spam. When unclear, it is treated as a lead.
  3. Leads & conversations (ManyChat): auto-replies with approved words only: the FAQ answer for a question, the out-of-hours holding line for a lead or problem, and your Calendly booking link if your routing rules say to offer one. Spam gets nothing.
  4. Leads & conversations (HubSpot): looks the person up first, then adds a note or creates one contact. Never a second record for the same person.
  5. Orchestrator again: wakes the owner only if it is real: an URGENT item at the top of queue/needs-your-ok.md. </aside>

What it reads

File Why it matters
brand-os/conversation/routing-rules.md What a real lead is, what is urgent, when to offer a booking link
brand-os/conversation/faq-library.md Approved answers, used word for word
brand-os/conversation/auto-reply-scripts.md The holding line and channel scripts
brand-os/brand-brain/voice-guide.md, tone-rules.md The voice check every reply passes

Who approves what

<aside> 🛑 Only words you approved in advance go out at night. These wait for you in queue/needs-your-ok.md:

A real lead or a problem goes to the top of the queue marked URGENT. Questions and spam never wake you.

</aside>

The command file

<aside> 📋 Copy this into .claude/commands/dm-lead-at-11pm.md.

</aside>

---
description: Playbook. A DM or lead comes in out of hours. Answers with approved words through ManyChat, logs the lead in HubSpot, and wakes the owner only if it is real.
argument-hint: "<paste the message: channel, sender name or handle, time and full text>"
---

# Playbook: DM or lead at 11pm

The event: $ARGUMENTS

## Before anything
1. If the file owner/STOP exists, stop. Append
   `YYYY-MM-DD HH:MM | orchestrator | stopped: owner/STOP present | -` to logs/activity.md
   and do nothing else.
2. Read owner/approval-bar.md, owner/brand-voice.md, owner/positioning.md and owner/budget-cap.md.
   They are read-only. If any still contains `[FILL IN:`, stop and tell the owner which file.

## Step 1. The orchestrator goes first
Use the orchestrator agent on this message as a playbook event
(priority: event, playbook: dm-lead-at-11pm). It writes one brief for
Leads & conversations to drafts/briefs/ with the message pasted in full and the time it arrived.

## Worth acting on
Every message gets an answer. Only some wake the owner.
Leads & conversations sorts it using the definitions in brand-os/conversation/routing-rules.md:
1. Real lead: someone asking to buy, book or get a quote.
2. Problem: a support problem, a complaint, anything about money, anything legal.
3. Question: answered word for word in brand-os/conversation/faq-library.md.
4. Spam: bots and noise.
When it is unclear, it is treated as a lead. If routing-rules.md defines none of these, do not
guess: treat every message as a lead, and the orchestrator adds an item to
queue/needs-your-ok.md asking the owner to fill routing-rules.md.

## Steps
1. Leads & conversations answers. Use the leads-conversations agent on its brief. It reads
   faq-library.md, routing-rules.md and brand-os/conversation/auto-reply-scripts.md, then:
   - Question: replies with the approved answer from faq-library.md, in the channel's
     script from auto-reply-scripts.md.
   - Real lead or problem: replies with the out-of-hours holding line from
     auto-reply-scripts.md.
   - Ready to talk, if routing-rules.md says so: adds the brand's Calendly booking link
     with the wording from auto-reply-scripts.md.
   - Spam: no reply.
   The reply goes out through ManyChat for Instagram and Facebook DMs. Until ManyChat is
   connected, the reply is written ready to paste. Every reply passes the voice check first;
   a fail means the person gets the holding line instead.
2. Leads & conversations records it. For a real lead, it looks the person up in HubSpot by
   email or phone first. If they exist, it adds a note. If not, it creates one contact with
   only what they wrote. Never a second record for the same person. Output:
   drafts/YYYY-MM-DD-inbox-<slug>.md.
3. Run the orchestrator again on the return. It applies the wake rule and the owner gate below.

## Wake the owner only if it is real
- Real lead or problem: the orchestrator puts an item at the top of the Open list in
  queue/needs-your-ok.md with URGENT in the title, in the format at the top of that file.
  This is the alert. The person already has the holding reply, so nothing else is sent.
- Question or spam: no alert. It shows up in the next session's summary.

## Who approves
These wait in queue/needs-your-ok.md:
- Any reply beyond the approved answer or holding line. New words never go out from here.
- Anything about a price, a discount, a refund or a date that is not word for word in
  faq-library.md.
- A complaint that may need a public reply. That goes to Reviews & reputation through the
  orchestrator.

## Log
Every agent appends one line per action to logs/activity.md:
`YYYY-MM-DD HH:MM | <agent> | <action> | <file>`
Example: `YYYY-MM-DD HH:MM | leads-conversations | holding reply sent, real lead logged in HubSpot | drafts/YYYY-MM-DD-inbox-<slug>.md`

## What the owner sees at the end
- Real lead or problem: an URGENT item at the top of the queue saying who wrote, what they
  asked, what was sent back and the HubSpot record
- Question or spam: one line in the next summary
- Any FAQ gap, so the owner can add the answer to faq-library.md

## Never
- Write new reply text and send it. Only approved answers and scripts go out.
- Quote, promise or hint at a price, discount, refund or delivery date.
- Wake the owner for spam or for a question the FAQ already answers.

Run it

<aside> ▶️ In Claude Code, from the project folder:

</aside>

/dm-lead-at-11pm "<paste the message: channel, sender, time and full text>"

<aside> 📄 What you get back: the sort, the reply that was sent (or written ready to paste until ManyChat is connected), the HubSpot record for a real lead, an URGENT queue item if it is real, and any FAQ gap so you can add the answer.

</aside>

Example

<aside> ☕ Example (Northwind Coffee, a fictional brand): late at night an Instagram DM asks whether Northwind can supply beans for a small café opening nearby.

<aside> 👉 Next: Brand OS: how the folder works

</aside>