pulse-content

name	pulse-content
version	2.0
type	autoresearch
description	Self-improving marketing content engine for Houston B2B
user-invocable	true
agent	PULSE
agent_model	google/gemini-flash-1.5
mcps	["paperclip","hermes","command-center"]
gstack_skills	["/design-consultation","/office-hours"]
eval_metric	lead_quality_score
eval_budget	60s
guard	brand_consistency > 0.8

/pulse-content — Self-Improving Content Engine

What it does

Generates and optimizes Houston B2B marketing content using Paperclip for competitive intelligence, Hermes for scheduled publishing, and gstack design skills for visual assets.

Protocol

Step 0: Load context

• memory/MARKETING_MEMORY.md — active campaigns, channel performance
• HENRY_BRAIN.md — brand positioning, ICP

Step 1: Competitive intelligence (Paperclip)

Fetch and analyze:

1. Competitor websites (Houston AI/automation companies)
2. Top-ranking pages for target SEO terms:
   • "Houston AI automation"
   • "Houston CPA AI"
   • "AI bookkeeping Houston"
3. LinkedIn content from similar agencies
4. Houston Business Journal AI coverage

Extract: messaging angles, content gaps, keyword opportunities.

Step 2: Content generation (PULSE + /office-hours)

For each content piece:

1. Run /office-hours to reframe the value proposition
2. Generate content in Whitt's voice (direct, Houston-local, outcome-first)
3. Types:
   • LinkedIn posts (3x/week) — 150 words max, hook + insight + CTA
   • Email sequences — pain → solution → proof → CTA
   • SEO blog outlines — long-tail Houston keywords
   • Fiverr/Upwork gig descriptions — service + outcome + proof

Step 3: Visual assets (/design-consultation)

For key pieces, run /design-consultation to generate:

• Social media graphics
• One-pager PDFs for prospects
• Email header images

Step 4: Schedule via Hermes

Use Hermes cron to:

• Queue LinkedIn posts at optimal times (Tue/Wed/Thu 8-10 AM CT)
• Schedule email sends
• Set content review reminders

Step 5: Track and optimize

Log to command-center:

• Impressions, engagement, click-through per piece
• Lead quality: did the content attract ICP-matching prospects?
• Feed winners back into content templates

Autoresearch Loop

• Metric: lead_quality_score — weighted score of leads generated (ICP match = 3, engagement = 2, conversion = 5)
• Budget: 60s eval
• Modify: content templates, hooks, CTAs, posting schedule
• Guard: brand_consistency stays above 0.8 (LLM judge on voice/tone match)

Copy Principles (from PULSE agent)

1. Lead with outcome, not feature
2. Speak to pain first
3. Social proof > features
4. One clear CTA per piece
5. Houston-specific language

MCP Integration Map

PULSE (Gemini Flash)
  ├── paperclip: competitive intel, SEO data, industry news
  ├── hermes: schedule publishing, set review reminders
  ├── command-center: track content performance metrics
  ├── /design-consultation: visual brand assets
  └── /office-hours: reframe value propositions