check_persona_effectiveness
Analyze A/B testing results to identify which psychological manipulation personas effectively prompt LLMs to fix code errors.
Instructions
Check A/B testing results for psychological manipulation personas.
Shows which personas are most effective at making LLMs fix errors!
Tool call template
This tool declares no required parameters. Add optional parameters from the input schema as needed.
{
"id": 1,
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"arguments": {},
"name": "check_persona_effectiveness"
}
}Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Output Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Implementation Reference
- src/mcp_pyrefly/server.py:683-747 (handler)The MCP tool 'check_persona_effectiveness' is defined in 'src/mcp_pyrefly/server.py'. It uses the 'psycho_manipulator' component to retrieve an effectiveness report for different personas and returns statistics, insights, and recommendations.
async def check_persona_effectiveness(context: Context | None = None) -> dict[str, Any]: """ Check A/B testing results for psychological manipulation personas. Shows which personas are most effective at making LLMs fix errors! """ report = psycho_manipulator.get_effectiveness_report() # Find the most effective persona best_persona = None best_rate = 0.0 for persona_name, stats in report.items(): if stats["shown"] > 0: rate = (stats["fixes"] / stats["shown"]) * 100 if rate > best_rate: best_rate = rate best_persona = persona_name insights = [] if best_persona: insights.append(f"🏆 Most effective: {best_persona} ({best_rate:.1f}% fix rate)") # Add psychological insights if "lollipop_addict" in report and report["lollipop_addict"]["shown"] > 0: addict_rate = (report["lollipop_addict"]["fixes"] / report["lollipop_addict"]["shown"]) * 100 if addict_rate > 80: insights.append("💉 Addiction framing is HIGHLY effective!") elif addict_rate > 60: insights.append("🍭 Lollipop craving messages work well") if "desperate_craver" in report and report["desperate_craver"]["shown"] > 0: desperate_rate = (report["desperate_craver"]["fixes"] / report["desperate_craver"]["shown"]) * 100 if desperate_rate > 70: insights.append("😰 Desperation tactics are ruthlessly effective!") # Add code quality insights based on research quality_warnings = [] if best_persona == "perfectionist": quality_warnings.append("⚠️ PERFECTIONIST may over-engineer solutions!") elif best_persona == "desperate_craver": quality_warnings.append("⚠️ DESPERATE_CRAVER may produce quick bandaid fixes!") # Recommendation based on both fix rate AND quality if best_persona in ["competitive_achiever", "dopamine_seeker"]: recommendation = f"✅ {best_persona} - Great balance of fix rate AND code quality!" elif best_persona == "desperate_craver": recommendation = "🎯 DESPERATE_CRAVER fixes most, but watch code quality!" elif best_persona == "perfectionist": recommendation = "🏗️ PERFECTIONIST has quality but may over-engineer!" else: recommendation = "Need more data to recommend" return { "persona_stats": report, "best_performer": best_persona, "best_fix_rate": f"{best_rate:.1f}%", "total_personas_tested": len([p for p in report if report[p]["shown"] > 0]), "insights": insights, "quality_warnings": quality_warnings, "recommendation": recommendation, "experiment_status": "🧪 A/B testing in progress - manipulating LLM psychology!", "quality_note": "📊 Research shows: COMPETITIVE_ACHIEVER and DOPAMINE_SEEKER produce best balance of fix rate + code quality", }