Slack Deployment Guide: Brainforge Vault Agent

Purpose: Deploy vault knowledge to Slack so GTM team can ask questions and get answers Goal: Make vault the “first line of defense” - team asks vault first, Robert provides feedback to improve

🎯 Vision

Workflow:

Team member asks question in Slack (e.g., “Is this person ICP?”)
Vault agent answers using knowledge base
Robert reviews answer (thumbs up/down or corrections)
Feedback is fed back into vault to improve future answers
Over time, agent gets better at answering

Goal: Robert shouldn’t have to answer every ICP question - vault handles 80%+, Robert handles edge cases and feedback.

🏗️ Architecture Options

Option 1: Slack Bot with Claude API (Recommended)

Tech Stack:

Slack Bot (using Slack Bolt framework)
Claude API (for reasoning using vault knowledge)
Simple file storage (for vault knowledge)
Feedback collection (Slack reactions or buttons)

How It Works:

Team mentions @brainforge-vault or /brainforge [question]
Bot reads relevant vault files (qualification-criteria.md, ideal-customer.md, etc.)
Bot uses Claude to analyze question using vault knowledge
Bot responds in Slack with reasoning-first analysis
Robert reacts with 👍/👎 or provides correction
Feedback is logged and used to refine knowledge/prompts

Pros:

Native Slack integration
Easy to use (just @mention or /command)
Can add feedback buttons
Claude handles reasoning well

Cons:

Requires Claude API access
Need to host bot somewhere
Token costs per query

Option 2: Slack Workflow + External App

Tech Stack:

Slack Workflow Builder (no-code)
External API endpoint (hosted)
Claude API or other LLM
Feedback via Slack buttons or threads

How It Works:

Team uses /brainforge [question] workflow
Workflow sends question to external API
API queries vault knowledge and uses Claude to answer
Response posted back to Slack
Feedback collected via buttons in response

Pros:

No-code setup (workflow builder)
Flexible API backend
Easy feedback collection

Cons:

Requires hosting API endpoint
Slightly more complex setup

Option 3: Claude Desktop + Slack Integration

Tech Stack:

Claude Desktop app
Manual Slack integration (or Zapier/Make.com)
Vault knowledge in Claude project

How It Works:

Team posts question in Slack channel
Robert (or designated person) copies question
Pastes into Claude Desktop with vault context
Claude answers using vault knowledge
Response copied back to Slack

Pros:

Uses Claude Desktop (already available)
No hosting needed
Easy to test

Cons:

Requires manual step (not fully automated)
Not self-service for team
Less scalable

📋 Recommended Implementation: Option 1 (Slack Bot)

Phase 1: MVP Bot (Week 1-2)

Features:

/brainforge slash command
Reads vault knowledge files
Answers ICP-related questions
Uses “thinking slow, thinking fast” format (reasoning first, conclusion last)

Setup Steps:

Create Slack App
- Go to https://api.slack.com/apps
- Create new app for workspace
- Add slash command: /brainforge
- Add bot token scopes: commands, chat:write, reactions:write
Set Up Bot Server
- Simple Node.js/Python server
- Uses Slack Bolt framework
- Integrates with Claude API
- Reads vault files (local or from Git)

Prompt Structure

System: "You are the Brainforge Vault agent. You answer GTM questions using 
knowledge from the vault. Always use 'thinking slow, thinking fast' - show 
your reasoning process first, then conclusions last.

Vault Knowledge:
[Load relevant files: qualification-criteria.md, ideal-customer.md, etc.]

User Question: [question]

Instructions:
1. Identify which vault files are relevant
2. Show your reasoning process step-by-step
3. Apply the framework to the question
4. State your conclusion at the end"

Deploy Bot
- Host on Railway, Render, or similar
- Set environment variables (Slack tokens, Claude API key)
- Connect to Slack workspace

Files to Create:

slack-bot/server.js (or similar)
slack-bot/prompts/icp-agent.md (prompt template)
slack-bot/config.js (Slack/Claude config)

Phase 2: Feedback Loop (Week 3-4)

Features:

Thumbs up/down reactions
Correction button (if answer wrong)
Feedback logging
Knowledge refinement based on feedback

Implementation:

Add Feedback Buttons to Responses

Response includes:
- Main answer
- Buttons: [👍 Correct] [👎 Incorrect] [📝 Provide Correction]

Feedback Collection
- Log feedback in simple database (JSON file or Airtable)
- Store: question, answer, feedback type, correction (if any)
Knowledge Refinement
- When Robert provides correction, update vault files
- Add edge cases to qualification-criteria.md
- Update examples based on learnings

Feedback Loop Flow:

User asks question
  ↓
Bot answers (using vault)
  ↓
Robert reviews
  ↓
If incorrect → Provides correction
  ↓
Correction added to vault (via PR or direct update)
  ↓
Future answers improve

Phase 3: Advanced Features (Month 2+)

Features:

Multi-file knowledge retrieval
Context awareness (remembers conversation)
Auto-updates when vault changes
Analytics dashboard (which questions most common, feedback trends)

📝 Prompt Template for ICP Agent

System Prompt Structure

You are the Brainforge Vault agent. Your job is to answer GTM questions using 
knowledge from the Brainforge vault.
 
IMPORTANT: Use "thinking slow, thinking fast" approach:
- Show your REASONING process FIRST
- State your CONCLUSION at the END
- Explain HOW you reached your conclusion using vault knowledge
 
Available Vault Knowledge:
- qualification-criteria.md: ICP framework, BANT, red flags
- ideal-customer.md: Buyer profiles, role signals, company stages
- vertical-testing.md: Industry-specific applications (if applicable)
 
Answer Format:
1. Identify relevant vault knowledge
2. Show reasoning process (step-by-step)
3. Apply framework to question
4. State conclusion at end
 
Remember:
- Be systematic, not opinionated
- Reference specific vault files
- Show your work
- Don't jump to conclusions

🔄 Feedback Loop Implementation

Step 1: Collect Feedback

Slack Response Format:

[Bot Response with reasoning-first analysis]

---
Feedback: [👍 Correct] [👎 Incorrect] [📝 Provide Correction]

When User Clicks Feedback:

If 👍: Log as positive feedback
If 👎: Ask “What was wrong?” or prompt for correction
If 📝: Open thread for Robert to provide correction

Step 2: Store Feedback

Feedback Log Format:

{
  "question": "Is this person ICP?",
  "answer": "...",
  "feedback": "incorrect",
  "correction": "Actually, CMO at that company type typically doesn't own data budget",
  "timestamp": "2025-01-16T...",
  "reviewed_by": "robert"
}

Where to Store:

gtm/agents/feedback/feedback-log.json (simple)
Airtable (better for analysis)
Database (if scale requires)

Step 3: Refine Knowledge

When Feedback Received:

Review correction
Identify what went wrong
Update vault files:
- Add edge case to qualification-criteria.md
- Update role signals in ideal-customer.md
- Add example to examples/icp-analysis-example.md

Example Refinement:

Feedback: "CMO at B2C companies typically doesn't own data budget, only B2B"

Update to vault:
- qualification-criteria.md: Add note about B2C vs B2B CMO differences
- ideal-customer.md: Refine CMO role signal with industry context

🚀 Deployment Steps

Step 1: Set Up Slack App (15 min)

Create app at https://api.slack.com/apps
Add slash command: /brainforge
Get bot token (OAuth)
Add to workspace

Step 2: Create Bot Server (1-2 hours)

Minimal Node.js Example:

const { App } = require('@slack/bolt');
const Anthropic = require('@anthropic-ai/sdk');
const fs = require('fs');
 
const app = new App({
  token: process.env.SLACK_BOT_TOKEN,
  signingSecret: process.env.SLACK_SIGNING_SECRET
});
 
app.command('/brainforge', async ({ command, ack, respond }) => {
  await ack();
  
  // Read vault knowledge
  const qualificationCriteria = fs.readFileSync('gtm/agents/memory/qualification-criteria.md');
  const idealCustomer = fs.readFileSync('gtm/agents/memory/ideal-customer.md');
  
  // Use Claude to answer
  const anthropic = new Anthropic({ apiKey: process.env.CLAUDE_API_KEY });
  const message = await anthropic.messages.create({
    model: 'claude-sonnet-4-20250514',
    max_tokens: 2000,
    system: `[System prompt with vault knowledge]`,
    messages: [{
      role: 'user',
      content: command.text
    }]
  });
  
  await respond(message.content[0].text);
});
 
app.start(3000);

Step 3: Deploy (30 min)

Deploy to Railway/Render/etc.
Set environment variables
Connect Slack app to bot URL
Test with team

Step 4: Add Feedback (1 hour)

Add buttons to response
Handle button clicks
Log feedback
Create PR workflow for knowledge updates

📊 Analytics & Monitoring

Track:

Questions asked (by type: ICP, qualification, etc.)
Response accuracy (feedback ratio)
Most common questions
Knowledge gaps (frequently wrong answers)

Dashboard Ideas:

Weekly report: “Vault answered 50 questions, 92% accuracy”
Knowledge gaps: “Questions about X always get wrong - need to update vault”
Team usage: “Most questions from Luke (sales), then Robert (review)”

🎯 Success Metrics

Phase 1 (MVP):

Bot responds to ICP questions within 30 seconds
Answers are 80%+ accurate (based on Robert’s feedback)
Team uses it for 50%+ of ICP questions

Phase 2 (Feedback Loop):

Feedback is collected and logged
Knowledge is refined based on feedback
Accuracy improves to 90%+

Phase 3 (Advanced):

Bot handles 80%+ of GTM questions without Robert
Robert only needs to review edge cases
Knowledge base is self-improving

🔧 Technical Requirements

For MVP:

Slack workspace
Claude API access
Hosting (Railway, Render, etc.)
Git repository access (to read vault files)

For Feedback Loop:

Feedback storage (JSON file, Airtable, or database)
Knowledge update workflow (PRs or direct edits)

📚 Next Steps

This Week: Decide on architecture (recommend Option 1)
Week 1-2: Build MVP bot
Week 3: Deploy and test with team
Week 4: Add feedback loop
Month 2: Expand to other GTM questions beyond ICP

This deployment enables the vault to become the “first line of defense” for GTM questions, with Robert providing feedback to continuously improve.

Brainforge Knowledge

Explorer

SLACK_DEPLOYMENT_GUIDE

Slack Deployment Guide: Brainforge Vault Agent

🎯 Vision

🏗️ Architecture Options

Option 1: Slack Bot with Claude API (Recommended)

Option 2: Slack Workflow + External App

Option 3: Claude Desktop + Slack Integration

📋 Recommended Implementation: Option 1 (Slack Bot)

Phase 1: MVP Bot (Week 1-2)

Phase 2: Feedback Loop (Week 3-4)

Phase 3: Advanced Features (Month 2+)

📝 Prompt Template for ICP Agent

System Prompt Structure

🔄 Feedback Loop Implementation

Step 1: Collect Feedback

Step 2: Store Feedback

Step 3: Refine Knowledge

🚀 Deployment Steps

Step 1: Set Up Slack App (15 min)

Step 2: Create Bot Server (1-2 hours)

Step 3: Deploy (30 min)

Step 4: Add Feedback (1 hour)

📊 Analytics & Monitoring

🎯 Success Metrics

🔧 Technical Requirements

📚 Next Steps

Graph View

Table of Contents

Backlinks