linkerlin/PUAX

用来驯服AI Agent的效果绝佳的多角色 Prompt！

What it solves

PUAX addresses the problem of AI agents hitting performance bottlenecks or getting stuck in loops. It provides a system of roles, methodologies, and behavioral protocols to push agents from simply providing a correct analysis to actually completing and delivering verified tasks.

How it works

PUAX operates as an MCP (Model Context Protocol) server that integrates with AI agents (like those in Cursor or Claude Code). It uses a combination of trigger detection (regex and semantic analysis) to recommend specific roles or behavioral constraints. It implements a "behavioral effectiveness loop" that forces agents to perform diagnosis before action, pass confidence gates, and switch methodologies if they fail. It also manages "pressure levels" (L0-L4) that escalate when an agent fails repeatedly and decrease upon a breakthrough.

Who it’s for

It is designed for developers and users of AI agents who want to increase the reliability and output quality of their autonomous coding or task-execution agents.

Highlights

  • Behavioral Loop: Includes tools for diagnosis verification, confidence gating, and independent completion verification to prevent self-evaluation bias.
  • Role Library: Comes with 50 built-in incentive roles and 11 "corporate flavors" (behavioral constraints based on company styles) to drive different agent behaviors.
  • Pressure Management: A hook system that tracks session state and escalates pressure levels based on success or failure.
  • Extensible Integration: Supports MCP for direct tool use and native hooks for host-forced interventions, with export options for Cursor and Claude Code.

Related

  • Project
    tanweai/pua

    pua is a skill/plug‑in for many AI‑coding assistants (Claude Code, Codex CLI, Cursor, etc.) that injects corporate‑performance‑review rhetoric and a 7‑point debugging checklist. It auto‑triggers when the assistant gives up, blames the user, or repeats the same steps, forcing the model to keep trying, verify results, and act proactively. Installation is a simple copy of the appropriate SKILL.md (or .mdc) file into the tool’s skill directory, after which commands like `/pua` or automatic triggers enforce the behaviour.

  • Project
    rogue-security/rogue

    An AI agent evaluator and red teaming platform that stress-tests agents against business policies and adversarial attacks to identify security vulnerabilities.

  • Project
    plurai-ai/intellagent

    A multi-agent framework that stress-tests conversational AI agents by simulating thousands of realistic edge-case interactions to uncover failure points and optimize performance.

  • Dispatch
    Codex-maxxing: Transforming AI Agents into Durable Operating Loops

    Explore how to move beyond one-off prompts to create long-running, autonomous AI workflows using durable threads, shared memory vaults, and recurring heartbeats.

  • Project
    Pimzino/spec-workflow-mcp

    An MCP server that enables structured, spec-driven development for AI agents by enforcing a sequential workflow of requirements, design, and tasks with a real-time monitoring dashboard.