ken-you-reflect

An MCP server that forces brutal honesty by challenging overconfident claims with uncomfortable questions.

Purpose

ken-you-reflect combats the tendency of LLMs (and developers) to claim everything is "perfect" or "working great" by forcing critical self-examination. It asks pointed questions that make you confront the real issues and risks in your code.

Features

• Brutal honesty questions - Forces you to confront what's actually wrong
• Pattern detection - Identifies toxic positivity patterns like "100% secure" or "zero bugs"
• Risk scoring - Automatically scores reflections based on severity (0-10)
• Context-rot detection - Helps LLMs recognize when old context is polluting current work
• Priority action phase - Prevents analysis paralysis by forcing THE ONE thing to fix
• Enhanced search - Filter by risk score, date, status to find critical issues

Installation

As an NPM package

npm install -g ken-you-reflect

As an MCP server

Add to your Claude Desktop config:

{
  "mcpServers": {
    "ken-you-reflect": {
      "command": "npx",
      "args": ["ken-you-reflect"]
    }
  }
}

Usage

Basic reflection

// In Claude or any MCP client
await use_mcp_tool("ken-you-reflect", "reflect", {
  claim: "Our authentication system is completely secure",
  context: "Just implemented JWT tokens with bcrypt",
});

Search reflections

// Find high-risk issues
await use_mcp_tool("ken-you-reflect", "search_reflections", {
  query: "security",
  minRisk: 8,
});

// Find recent issues
await use_mcp_tool("ken-you-reflect", "search_reflections", {
  query: "authentication",
  daysAgo: 7,
  status: "pending",
});

Check context (v1.1.0)

// Check if your context is polluted
await use_mcp_tool("ken-you-reflect", "check_context", {
  current_task: "Fix authentication bug",
  concerns: "Been jumping between 3 different features",
});

Tools

1. reflect - Challenge a claim with brutal honesty questions
2. update_reflection - Mark a reflection as addressed or accepted-with-risks
3. list_reflections - List reflections with optional status filter
4. search_reflections - Search with filters for risk, date, and status
5. record_assessment - Record your brutal assessment and critical issues
6. check_context - Check if your context is helping or hurting your current task (v1.1.0)

Question Examples

• "Stop. Is what you're building RIGHT NOW moving toward what they actually asked for?"
• "If someone's job depended on this being perfect, what would they panic about?"
• "Stop. Is your current context helping or hurting your ability to complete THIS specific task?"
• "Would you bet your job that this will work perfectly in production?"

Pattern Detection

Automatically detects dangerous claim patterns:

• absolute-positive: "perfect", "flawless", "100%"
• zero-defect-claim: "no bugs", "zero issues"
• completion-claim: "done", "complete", "ready"

Integration with Hooks

Create shell hooks that trigger on positive claims:

#!/bin/bash
# example-hook.sh
if [[ "$1" =~ (perfect|flawless|complete|secure|ready) ]]; then
  echo "⚠️ Overconfident claim detected!"
  exit 2  # Triggers reflection
fi

License

MIT

Author

KenKaiii [email protected]