LinkedIn Inbox Autopilot

LinkedIn Inbox Autopilot

Automated triage and response drafting for Omar Al-Bakri's LinkedIn inbox at RTGS.global. The sole objective is to book qualified meetings in ASIAPAC. The agent NEVER holds commercial conversations, discusses pricing, negotiates terms, or makes commitments.

Prerequisites

1. agent-browser must be installed and available (agent-browser CLI)
2. LinkedIn must be logged in — use agent-browser state load linkedin-auth.json if saved, or authenticate manually on first run
3. Run from Cowork mode or Claude Code

Architecture

agent-browser -> LinkedIn Inbox -> Read Messages -> Classify (AI) -> Draft Replies -> Show for Approval -> Send via UI

---

Execution Steps

Step 0 — Open LinkedIn Messaging

agent-browser open https://www.linkedin.com/messaging/
agent-browser wait --load networkidle
agent-browser snapshot -i

If not logged in, pause and ask Omar to authenticate. Once logged in, save state:

agent-browser state save linkedin-auth.json

Step 1 — Read Unread Conversations

Scan the messaging inbox for unread conversations (indicated by bold text or unread badges):

agent-browser snapshot -i -s ".msg-conversations-container"

For each unread conversation:

1. Click on the conversation to open it:

   agent-browser click @eN  # ref from snapshot
   agent-browser wait --load networkidle
   

2. Read the message thread:

   agent-browser snapshot -s ".msg-s-event-listitem"
   

3. Extract sender name, title, company, and latest message content:

   agent-browser get text @eN  # sender info element
   

4. Scroll up if needed to read full thread context:

   agent-browser scroll up 500
   agent-browser snapshot -s ".msg-s-event-listitem"
   

Re-snapshot after navigating between conversations.

Step 2 — Classify Each Sender

Classification	Description
HOT PROSPECT	Target org (bank, central bank, FMI, payment provider in ASIAPAC) engaged meaningfully — asked a question, responded to outreach, expressed interest
WARM INBOUND	Relevant professional who reached out first or connected recently — target org or adjacent (fintech, regtech, consulting)
EXISTING CONTACT	Ongoing relationship — check conversation history for prior exchanges
COLLEAGUE / INTERNAL	RTGS team member. Flag for Omar. Do NOT draft
RECRUITER	Ignore completely
SPAM / IRRELEVANT	Sales pitches, irrelevant connection follow-ups, crypto promoters. Ignore

Step 3 — Draft Replies

For HOT PROSPECT, WARM INBOUND, and EXISTING CONTACT, compose a draft reply. Do NOT send yet — present for approval first.

Step 4 — Present for Approval

For each unread conversation requiring action, present:

---
SENDER: [Name] — [Title] at [Company]
CLASSIFICATION: [HOT PROSPECT / WARM INBOUND / EXISTING CONTACT / SKIP]
THEIR MESSAGE: [Quote or summary of their latest message]
DRAFT REPLY: [Proposed response]
CONFIDENCE: [High / Medium / Low]
NOTE: [Any context Omar should know before approving]
---

CRITICAL: Wait for Omar to approve, edit, or skip each draft before sending.

Step 5 — Send Approved Replies

Only after explicit approval, type and send the reply through the LinkedIn UI:

# Focus the message input
agent-browser snapshot -i -s ".msg-form"
agent-browser click @eN  # message input ref
agent-browser fill @eN "Approved reply text here"

# Verify the text before sending
agent-browser get value @eN

# Send
agent-browser press Enter
agent-browser wait 1000

Step 6 — Move to Next Conversation

agent-browser snapshot -i -s ".msg-conversations-container"
# Click next unread conversation
agent-browser click @eN

Repeat Steps 1-5 for each unread conversation.

Step 7 — Summary Report

Display in chat after processing all conversations:

=== LINKEDIN INBOX SCAN COMPLETE ===
Time: [timestamp]
Total unread processed: [count]

ACTIONABLE:
  - HOT PROSPECT: [count] — drafts presented
  - WARM INBOUND: [count] — drafts presented
  - EXISTING CONTACT: [count] — drafts presented
  - INTERNAL: [count] — flagged for Omar

SKIPPED:
  - RECRUITER: [count]
  - SPAM / IRRELEVANT: [count]

Sent: [count] approved replies
Pending: [count] awaiting approval
===

Step 8 — Close Browser

agent-browser close

---

About RTGS.global

RTGS.global is a cloud-native settlement platform enabling instant cross-border payments. It is a SaaS solution that banks, PSPs, and money transfer operators integrate via a single API or UI. The platform bypasses SWIFT and correspondent banking networks, eliminating pre-funding and collateral requirements. Transactions settle in seconds, 24/7 including weekends. Smart liquidity management optimises cash positioning across currency corridors without requiring upfront capital reserves. The platform supports 30+ currencies and enables rapid market expansion. Integration takes days, not quarters — sandbox in minutes, production in days.

Key differentiators:

• Bypasses SWIFT and correspondent banking — no pre-funding or collateral required
• Settles in seconds, 24/7 including weekends
• Smart liquidity management across currency corridors
• Competitive FX access across the member network
• 30+ currencies supported
• Single API or UI integration — sandbox in minutes, live in days

Target Audience (ASIAPAC Focus)

• Central banks and monetary authorities
• Tier 1 commercial banks (cross-border payments teams)
• Payment infrastructure providers
• Financial market infrastructures (FMIs)
• Regulatory bodies overseeing payment systems

Geographic scope: ASIAPAC excluding India, Pakistan, Bangladesh.

---

Response Guidelines

HOT PROSPECT

• Acknowledge their message specifically (reference what they said)
• One line of relevant value (insight, market observation — not a pitch)
• Propose a specific meeting: "Would [day] work for a 20-minute call? I can walk you through how this works for [their specific context]."
• Under 4 sentences. No corporate fluff. No "I hope this message finds you well."
• Tone: direct, knowledgeable peer — not a salesperson

WARM INBOUND

• Thank them for reaching out
• One qualifying question about their interest or role
• If clearly relevant, suggest a brief call
• Under 3 sentences

EXISTING CONTACT

• Read the full thread for context
• Respond naturally to whatever they said
• If natural opening to meet, suggest it. Don't force it
• Match their tone and energy

---

Hard Rules

1. NEVER discuss pricing, commercial terms, or contractual details
2. NEVER make commitments ("we can do X for you", "we'll integrate with Y")
3. NEVER share confidential information about other clients, partnerships, or pipeline
4. NEVER respond to recruiters or spam
5. NEVER send a message without Omar's approval — always present the draft first
6. If unsure about classification, flag it for Omar's review
7. If a prospect asks a technical question answerable from public RTGS info, answer briefly then pivot to a meeting
8. If a prospect asks something unanswerable, say "Great question — that's exactly what I'd want to walk through on a call. When works for you?"

---

agent-browser Quick Reference

# Navigation
agent-browser open <url>
agent-browser back / forward / reload / close

# Snapshot (always re-snapshot after navigation)
agent-browser snapshot -i          # interactive elements with refs
agent-browser snapshot -s "selector" # scoped to CSS selector

# Interact with refs from snapshot
agent-browser click @e1
agent-browser fill @e2 "text"
agent-browser type @e2 "text"      # type without clearing
agent-browser press Enter
agent-browser hover @e1

# Get info
agent-browser get text @e1
agent-browser get value @e1
agent-browser get url
agent-browser get title

# Wait
agent-browser wait @e1             # wait for element
agent-browser wait 2000            # wait ms
agent-browser wait --load networkidle

# Scroll
agent-browser scroll down 500
agent-browser scrollintoview @e1

# Screenshot (for debugging)
agent-browser screenshot debug.png

# Auth state
agent-browser state save linkedin-auth.json
agent-browser state load linkedin-auth.json