closer-autopilot

name	closer-autopilot
version	2.0
type	autoresearch
description	Autonomous prospect research and sales outreach optimization
user-invocable	true
agent	CLOSER
agent_model	google/gemini-flash-1.5
mcps	["paperclip","hermes","command-center"]
gstack_skills	["/office-hours"]
eval_metric	outreach_response_rate
eval_budget	60s
guard	false_positive_rate < 0.15

/closer-autopilot — Autonomous Sales Pipeline

What it does

Automates the CLOSER agent's prospect research → outreach → follow-up cycle using Paperclip for web intelligence, Hermes for scheduling, and /office-hours for pitch reframing.

Protocol

Step 0: Load context

Read these files for current state:

• memory/SALES_MEMORY.md — pipeline, active prospects, win/loss history
• memory/HENRY_CONTEXT.md — current priorities
• HENRY_BRAIN.md — target ICP and deal criteria

Step 1: Prospect research (Paperclip)

For each prospect in the pipeline or new target:

1. Fetch company website via Paperclip MCP

   • Extract: services offered, team size, owner name/age, tech stack
   • Flag: retirement indicators, manual processes, growth/decline signals

2. Fetch public profiles via Paperclip

   • LinkedIn (public view), Google reviews, TX State Board records
   • Note: confidence level (high/medium/low) on each data point

3. Score prospect against ICP:

   • Houston CPA/legal/medical firm? (+3)
   • Owner age 45-65? (+2)
   • 5-50 employees? (+2)
   • Manual back-office processes visible? (+3)
   • Total score /10 — qualify at 6+

Step 2: Pitch customization (/office-hours)

Run /office-hours with this prompt for each qualified prospect:

I'm pitching AI transformation to [COMPANY NAME], a [SIZE] [INDUSTRY] firm in Houston. Owner is [NAME], approximately [AGE]. They currently [PAIN POINTS observed from research]. My offer is a free 30-min AI audit — show them one thing their back office could automate TODAY. Help me reframe the approach for this specific prospect.

Capture the reframed pitch angle.

Step 3: Outreach generation

For each qualified prospect, generate:

1. LinkedIn DM (< 300 chars)

   • Lead with their specific pain point
   • Offer: "I built something that does [X] automatically — want to see it?"
   • No pitch, no sell — just curiosity

2. Cold email (< 150 words)

   • Subject: specific to their business, not generic
   • Body: outcome-first, one clear CTA
   • Sign-off: Whitt, HENRY AI Corporation

3. Follow-up sequence (Day 1, 3, 7)

   • Each adds new value, not "just checking in"

Step 4: Schedule via Hermes

Use Hermes MCP to:

• Schedule outreach sends at optimal times (Tue-Thu, 9-11 AM CT)
• Set follow-up reminders on Day 1/3/7
• Log all activity to command-center

Step 5: Track and optimize

After each outreach batch:

• Log response rate to results.tsv
• Analyze: which pitch angles got responses?
• Feed wins back into Step 2 for next iteration

Autoresearch Loop

This skill improves itself by tracking outreach_response_rate:

• Metric: % of outreach that gets a reply (any reply = counted)
• Budget: 60s eval per iteration
• Modify: pitch templates, subject lines, DM copy, timing
• Guard: false positive rate stays below 15% (don't spam unqualified leads)

MCP Integration Map

CLOSER (Gemini Flash)
  ├── paperclip: fetch prospect data from web
  ├── hermes: schedule sends + follow-ups via cron
  ├── command-center: log activities + track pipeline
  └── /office-hours: reframe pitch per prospect

Current Pipeline Targets

Target	Revenue	Status	Priority
TXS5345	$142K	LOI ready	P0
TXS5513	$424K	Awaiting buyer package	P0
TXS5491	$910K	Sourcing	P1
TXS5450	$472K	Research	P1
Houston CPAs (20)	Agency leads	Not started	P0