prompt-analyser/README.md

Prompt Performance Analytics

A two-agent system for analyzing, scoring, and improving AI prompts — for both humans and enterprise multi-agent systems.

Quick Start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Set up Anthropic API key
cp .env.example .env
# Edit .env with your Anthropic API key

# 3. Run the server
uvicorn backend.main:app --reload --port 8000

# 4. Open in browser
# Analyzer:  http://localhost:8000/
# Dashboard: http://localhost:8000/dashboard-ui

Architecture

Agent 1: Prompt Analyzer ──→ Agent 2: Analytics Reporter ──→ Dashboard
    ↑                              ↑
    ├── REST API (humans)          └── SQLite DB
    └── MCP Server (agents)

Agent 1: Prompt Analyzer

Scores prompts on 5 dimensions (clarity, token efficiency, goal alignment, structure, vagueness), identifies mistakes, and generates optimized rewrites. Uses Anthropic Claude via the official SDK.

Agent 2: Analytics Reporter

Aggregates all analyses into trends, mistake frequencies, and agent rankings, then serves data to the dashboard.

Context Store

Per-project isolated memory. Each project's history, patterns, and agent profiles are stored separately — no cross-contamination.

API Endpoints

Method	Endpoint	Description
POST	/analyze	Analyze a prompt
POST	/rewrite-choice	Record rewrite acceptance
GET	/dashboard/overview	KPI overview
GET	/dashboard/interactions	Paginated interaction feed
GET	/dashboard/trends?days=N&hours=N	Quality score trends
GET	/dashboard/mistakes	Common mistake types
GET	/dashboard/agents	Agent leaderboard

MCP Server (for agent-to-agent)

python -m mcp_server.server

Tools exposed:

• analyze_prompt — analyze a prompt with optional project context
• get_analysis_history — retrieve past analyses

Environment Variables

Variable	Description	Default
ANTHROPIC_API_KEY	Anthropic API key	—
ANTHROPIC_MODEL	Claude model to use	claude-sonnet-4-20250514
LLM_MAX_TOKENS	Max output tokens	4096
LLM_TEMPERATURE	Generation temperature	0.3