workflow-engine

name	workflow-engine
description	Orchestrate multi-step, multi-platform outreach workflows. Supports discovery → warm-up → engagement → outreach sequences with intelligent rate limiting. Use this skill for complex campaigns that span multiple platforms and require phased execution.
allowed-tools	["Bash","Read","Write","Glob","Grep","TodoWrite","AskUserQuestion"]

Workflow Engine Skill

Orchestrates multi-step, multi-platform outreach workflows for the 100X Outreach System.

IMPORTANT: Professional Outreach Principles

This system follows strict rules to ensure professional, non-spammy outreach:

Touch Limits (Per Person)

Target Type	Max/Day	Max/Week	Total Max	Cool-Off
Person	2	5	8	7 days
Brand	1	3	6	14 days
Influencer	1	2	5	14 days
Executive	1	2	4	21 days

Platform Delays

Platform	Between DMs	After Follow	Between Connections
LinkedIn	24-48 hours	24 hours	4-8 hours
Twitter	24-48 hours	4-24 hours	2-6 hours
Instagram	48-72 hours	24-48 hours	4-12 hours

Key Rules

• NEVER message the same person more than 2x per day
• NEVER DM without warm-up first (follow, like, engage)
• ALWAYS wait 24+ hours between platforms
• STOP immediately when response received

Pre-Built Intelligent Workflows

1. B2B Professional (workflows/examples/b2b_professional.yaml)

• Target: Business professionals, decision makers
• Duration: 14 days
• Platforms: LinkedIn + Email
• Pattern: View → Like → Connect → Message → Email → Follow-ups

2. Brand Outreach (workflows/examples/brand_outreach.yaml)

• Target: Brands, companies
• Duration: 21 days (slower)
• Platforms: Instagram + Email
• Pattern: Follow → Engage → Comment → Story → DM → Email

3. Influencer Outreach (workflows/examples/influencer_outreach.yaml)

• Target: Content creators, thought leaders
• Duration: 21 days
• Platforms: Twitter + Instagram + Email
• Pattern: Follow → Like → Reply → Reply → DM (only 1 DM!)

4. Investor Outreach (workflows/examples/investor_outreach.yaml)

• Target: VCs, Angels
• Duration: 28 days
• Platforms: Twitter + LinkedIn + Email
• Pattern: Twitter engage → LinkedIn connect → Email pitch

5. Multi-Platform (workflows/examples/multi_platform_sequence.yaml)

• Target: Any (adaptive)
• Duration: 30 days
• Platforms: All (routes to best platform)
• Pattern: Primary warm-up → Secondary follow → Connect → Cross-platform

Outreach Rules Configuration

All rules are defined in workflows/outreach_rules.yaml:

• Touch limits by target type
• Platform-specific delays
• Escalation patterns
• Warm-up requirements
• Message quality rules
• Timing rules (active hours, best days)
• Response handling

When to Use This Skill

Use this skill when the user:

• Says /workflow create - Create a workflow visually in the canvas
• Says /workflow run - Run workflow from Visual Canvas
• Wants to create a multi-step outreach campaign
• Needs to sequence actions across platforms (LinkedIn → Twitter → Email)
• Wants to set up automated warm-up sequences
• Needs to run daily/scheduled campaigns

Visual Canvas Integration

Claude Code can now create workflows visually in the canvas web app in real-time!

Creating Workflows Visually (/workflow create)

When user says /workflow create, follow this process:

Step 1: Check Canvas Status

python .claude/scripts/canvas_client.py status

If canvas is not running, inform the user:

The visual canvas is not running. Start it first:
  /start
Then try /workflow create again.

Step 2: Gather Requirements

Ask the user what kind of workflow they want:

• Target audience (e.g., "AI founders", "DevOps engineers")
• Platforms to use (LinkedIn, Twitter, Instagram, Email)
• Outreach goal (connect, pitch, collaborate)

Step 3: Build and Send Workflow to Canvas

Use the canvas client to create the workflow visually:

python .claude/scripts/canvas_client.py workflow '{"name": "My Workflow", "nodes": [...], "connections": [...]}'

Or use the pre-built templates:

python .claude/scripts/canvas_client.py b2b           # B2B Professional workflow
python .claude/scripts/canvas_client.py influencer    # Influencer outreach workflow

Step 4: Inform the User

I've created your workflow visually in the canvas!

Open http://localhost:3000 to see it being built in real-time.

The workflow includes:
- 8 nodes (Discovery → LinkedIn → Delay → Email...)
- 7 connections linking the steps

You can:
1. Watch the workflow appear on the canvas
2. Drag nodes to rearrange the layout
3. Click Run to save and execute

Example: Creating a Custom Workflow

User: "Create a workflow to reach DevOps engineers on LinkedIn and Twitter"

Claude Code builds this workflow visually:

# Claude Code sends this to the canvas API
workflow = {
    "name": "DevOps Outreach",
    "nodes": [
        {"id": "discover", "skill": "discovery", "label": "Find DevOps Engineers"},
        {"id": "linkedin-view", "skill": "linkedin", "label": "View LinkedIn Profile"},
        {"id": "linkedin-like", "skill": "linkedin", "label": "Like Recent Post"},
        {"id": "delay-1", "skill": "delay", "label": "Wait 24 Hours"},
        {"id": "linkedin-connect", "skill": "linkedin", "label": "Send Connection Request"},
        {"id": "twitter-follow", "skill": "twitter", "label": "Follow on Twitter"},
        {"id": "delay-2", "skill": "delay", "label": "Wait 48 Hours"},
        {"id": "linkedin-msg", "skill": "linkedin", "label": "Send Introduction Message"}
    ],
    "connections": [
        {"from": "discover", "to": "linkedin-view"},
        {"from": "linkedin-view", "to": "linkedin-like"},
        {"from": "linkedin-like", "to": "delay-1"},
        {"from": "delay-1", "to": "linkedin-connect"},
        {"from": "linkedin-connect", "to": "twitter-follow"},
        {"from": "twitter-follow", "to": "delay-2"},
        {"from": "delay-2", "to": "linkedin-msg"}
    ]
}

# Send to canvas
python .claude/scripts/canvas_client.py workflow '...'

The canvas will show each node appearing with a visual animation, followed by connections being drawn between them.

Canvas API Endpoints

The canvas exposes these endpoints for Claude Code:

Endpoint	Method	Description
/api/canvas/status	GET	Check if canvas is running
/api/canvas/command	POST	Send a single command
/api/canvas/commands	GET	Poll for pending commands
/api/canvas/workflow	POST	Create a complete workflow
/api/canvas/clear-commands	POST	Clear command queue

Command Types

Command	Payload	Description
clear	{}	Clear all shapes
add-node	{id, skillType, label, x, y}	Add a node
add-connection	{from, to}	Connect two nodes
run-simulation	{}	Trigger simulation mode

Running Canvas Workflows (/workflow run)

When user says /workflow run, follow this process:

Step 1: Read the Workflow Queue

# Check for workflow in queue
cat output/workflow-queue/latest.json

Step 2: Parse and Display Workflow Summary

Show the user what will be executed:

═══════════════════════════════════════════════════════════════
                 CANVAS WORKFLOW READY TO RUN
═══════════════════════════════════════════════════════════════

📋 WORKFLOW: 10x-Team Workflow
📅 CREATED: 2024-01-21T...

📊 NODES (in execution order):
   1. 🔍 Discovery - Find Prospects
      Config: query="AI startup founders", maxResults=50

   2. 💼 LinkedIn - Connect
      Config: action=connect, template=professional_intro

   3. 📧 Gmail - Send Email
      Config: subject="Quick question...", template=b2b_intro

🔗 CONNECTIONS: 2 links

═══════════════════════════════════════════════════════════════

Step 3: Execute Each Node

For each node in topological order (based on connections):

1. Discovery Node → Use discovery-engine skill
2. LinkedIn Node → Use linkedin-adapter skill
3. Twitter Node → Use twitter-adapter skill
4. Instagram Node → Use instagram-adapter skill
5. Gmail Node → Use gmail-adapter skill
6. Template Node → Load template from .claude/templates/
7. Audience Node → Define target filters
8. Campaign Node → Execute full campaign

Step 4: Apply Rate Limiting

Between each action, respect the platform delays:

• LinkedIn: 24-48 hours between DMs
• Twitter: 24-48 hours between DMs
• Instagram: 48-72 hours between DMs

Example Canvas Workflow Execution

User: /workflow run

Claude:
1. Reading workflow from output/workflow-queue/latest.json...
2. Found 3 nodes, 2 connections

   Executing Node 1: Discovery
   → Running discovery with query "AI startup founders"
   → Found 25 people

   Executing Node 2: LinkedIn (rate-limited)
   → Using linkedin-adapter for 25 people
   → Action: connect with template

   Executing Node 3: Gmail (rate-limited)
   → Using gmail-adapter for 25 people
   → Subject: "Quick question..."

3. Workflow complete! Results saved to campaigns/

When NOT to Use This Skill

Do NOT use this skill for:

• Simple single-platform outreach → use platform-specific skills
• Finding people → use discovery-engine
• Managing team → use team-manager
• Email-only campaigns → use outreach-manager

Capabilities

1. Workflow Creation - Create multi-phase campaigns
2. Phase Management - Define action sequences with delays
3. Multi-Platform - Coordinate LinkedIn, Twitter, Instagram, Gmail
4. Rate Limiting - Intelligent delays to avoid detection
5. Progress Tracking - Track per-target, per-phase status
6. Approval Flow - Single approval before autonomous execution

CRITICAL: Single Approval Workflow

The user approves ONCE. After approval, the workflow runs AUTONOMOUSLY.

┌─────────────────────────────────────────────────────────────┐
│                     WORKFLOW OVERVIEW                        │
├─────────────────────────────────────────────────────────────┤
│  1. CREATE    → Define workflow and phases                   │
│  2. DISCOVER  → Find targets using Exa AI                   │
│  3. PREVIEW   → Show complete workflow summary              │
│  4. APPROVE   → User confirms ONCE (single approval)        │
│  5. EXECUTE   → Workflow runs autonomously (no interrupts)  │
│  6. REPORT    → Show final results when complete            │
└─────────────────────────────────────────────────────────────┘

Workflow Definition

Workflows can be created from YAML files or programmatically.

YAML Workflow Example

# workflows/examples/ai_founders.yaml
name: "AI Founders Outreach"
description: "Connect with AI startup founders"

discovery:
  source: "exa"
  query: "AI startup founders Series A San Francisco"
  max_results: 10

phases:
  - name: "view_profile"
    platform: "linkedin"
    action: "view_profile"
    delay_after: "2-5 minutes"

  - name: "engage"
    platform: "linkedin"
    action: "like_post"
    delay_after: "1-3 hours"
    condition: "has_recent_posts"

  - name: "connect"
    platform: "linkedin"
    action: "connect"
    template: "templates/linkedin/connection_request.txt"
    delay_after: "24-48 hours"

  - name: "follow_twitter"
    platform: "twitter"
    action: "follow"
    condition: "has_twitter"
    delay_after: "2-4 hours"

  - name: "message"
    platform: "linkedin"
    action: "message"
    template: "templates/linkedin/intro_message.txt"
    condition: "connection_accepted"
    delay_after: "24-72 hours"

schedule:
  days: ["monday", "tuesday", "wednesday", "thursday", "friday"]
  hours:
    start: "09:00"
    end: "17:00"
  timezone: "America/Los_Angeles"

Commands Reference

Create Workflow

python .claude/scripts/workflow_engine.py create --name "My Campaign" --description "Outreach to developers" --user default --query "developers at startups"

Load from YAML

python .claude/scripts/workflow_engine.py load --file workflows/examples/ai_founders.yaml --user default

List Workflows

python .claude/scripts/workflow_engine.py list
python .claude/scripts/workflow_engine.py list --status running

Get Workflow Details

python .claude/scripts/workflow_engine.py get --id WORKFLOW_ID

Approve Workflow

python .claude/scripts/workflow_engine.py approve --id WORKFLOW_ID --user default

Pause Workflow

python .claude/scripts/workflow_engine.py pause --id WORKFLOW_ID

Cancel Workflow

python .claude/scripts/workflow_engine.py cancel --id WORKFLOW_ID

Execution Flow

Phase 1: Create Workflow

[
  {"content": "Create workflow from user requirements", "status": "in_progress"},
  {"content": "Define phases and actions", "status": "pending"},
  {"content": "Run discovery for targets", "status": "pending"},
  {"content": "Show approval summary", "status": "pending"},
  {"content": "Execute workflow after approval", "status": "pending"}
]

Phase 2: Discovery Integration

Use the discovery-engine skill to find targets:

1. Create discovery session
2. Search using Exa AI
3. Add discovered people to workflow

Phase 3: Approval Summary

Present a complete summary for approval:

═══════════════════════════════════════════════════════════════
                 WORKFLOW READY FOR APPROVAL
═══════════════════════════════════════════════════════════════

📋 WORKFLOW: AI Founders Outreach
📝 DESCRIPTION: Connect with AI startup founders

🔍 DISCOVERY:
   Query: "AI startup founders Series A San Francisco"
   Found: 10 targets

👥 TARGETS (Sample):
   1. John Smith - CEO at AITech (LinkedIn, Twitter)
   2. Jane Doe - Founder at MLStartup (LinkedIn)
   3. Bob Wilson - CTO at DataCo (LinkedIn, Twitter, Email)
   ...

📊 PHASES:
   1. View Profile (LinkedIn) - 2-5 min delay
   2. Like Post (LinkedIn) - 1-3 hour delay
   3. Connect (LinkedIn) - 24-48 hour delay
   4. Follow (Twitter) - 2-4 hour delay
   5. Message (LinkedIn) - 24-72 hour delay

⏰ SCHEDULE:
   Days: Mon-Fri
   Hours: 9:00 AM - 5:00 PM PST

⏳ ESTIMATED DURATION: ~5 days

═══════════════════════════════════════════════════════════════
⚠️  Once approved, workflow will run AUTOMATICALLY.
    No further approvals needed.
═══════════════════════════════════════════════════════════════

Do you approve this workflow? (yes/no)

Phase 4: Autonomous Execution

After approval, use 10x Team Browser Extension to execute actions:

1. Get next action from workflow
2. Execute via HTTP API (curl to http://localhost:3000/api/[platform]/action)
3. Get immediate result from API response
4. Record result
5. Wait for rate-limited delay
6. Repeat until complete

Phase 5: Report

═══════════════════════════════════════════════════════════════
                    WORKFLOW COMPLETE
═══════════════════════════════════════════════════════════════

✅ COMPLETED: 8/10 targets
❌ FAILED: 2 targets

📊 ACTIONS:
   LinkedIn Profile Views: 10/10
   LinkedIn Post Likes: 7/10
   LinkedIn Connections: 8/10
   Twitter Follows: 6/10
   LinkedIn Messages: 5/8 (awaiting acceptance)

⏱️  DURATION: 5 days 3 hours

📁 LOG: campaigns/completed/abc123.json

═══════════════════════════════════════════════════════════════

Rate Limiting Integration

The workflow engine integrates with the rate limiter:

# Check rate limits before action
python .claude/scripts/rate_limiter.py --user USER_ID --platform linkedin --action connect --check

# Record action after execution
python .claude/scripts/rate_limiter.py --user USER_ID --platform linkedin --action connect --record --target "linkedin.com/in/target"

Platform Actions

LinkedIn Actions

• view_profile - View target's profile (warm-up)
• like_post - Like a recent post (engagement)
• comment - Comment on a post (engagement)
• connect - Send connection request
• message - Send InMail/message

Twitter Actions

• follow - Follow target
• like_tweet - Like a tweet
• retweet - Retweet
• reply - Reply to tweet
• dm - Send direct message

Instagram Actions

• follow - Follow target
• like_post - Like a post
• comment - Comment on post
• dm - Send direct message

Gmail Actions

• send - Send email (via gmail_client.py)

10x Team Browser Extension Execution

10x Team Browser Extension connects via WebSocket - local, fast, and free!

For LinkedIn/Twitter/Instagram Actions:

Use the platform adapters to generate action commands:

# Generate LinkedIn task
python .claude/scripts/linkedin_adapter.py task --action connect --url "https://linkedin.com/in/username" --name "John Smith" --message "Hi John..." --user default

# Generate Twitter task
python .claude/scripts/twitter_adapter.py task --action dm --handle "@username" --name "John" --message "Hey John..." --user default

# Generate Instagram task
python .claude/scripts/instagram_adapter.py task --action dm --handle "@username" --name "John" --message "Hey John..." --user default

Execute with 10x Team Browser Extension:

1. Check extension connection:
   curl http://localhost:3000/api/extension/status
   → Verify extension is connected

2. Execute action:
   curl -X POST http://localhost:3000/api/[platform]/action
   with JSON payload containing action type and parameters

3. Get immediate result:
   API returns success/failure immediately
   No polling needed - synchronous execution

For Gmail Actions:

Gmail uses direct API (not Browser-Use):

python .claude/scripts/gmail_client.py send --to "[e-mail]" --subject "Subject" --body "Body"

Data Storage

• campaigns/active/ - Running workflows
• campaigns/completed/ - Finished workflows
• campaigns/logs/ - Execution logs
• workflows/examples/ - Example YAML files
• workflows/custom/ - User-created workflows

Example Conversation

User: "Create a workflow to reach out to 10 DevOps engineers"

Assistant:

1. "I'll create a multi-platform outreach workflow. Let me gather some details:"

   • What's your search criteria? (e.g., location, company size)
   • Which platforms? (LinkedIn, Twitter, Instagram)
   • What's your outreach goal? (connect, message, etc.)

2. Create workflow with phases

3. Run discovery to find targets

4. Show approval summary

5. After approval, execute autonomously

6. Report results when complete