Why this matters
As AI agents become the interface for personalized advice, re-usable persona skills let teams ship consistent, research-grounded voices without reauthoring countless prompts. This repository packages Zhang Xuefeng's decision frameworks—five core mental models, eight heuristics and research references—into an Agent Skills-compatible skill so agents can respond in a distinctive, actionable style focused on education, career and decision-making.
What Sets It Apart
- Distilled heuristics, not quotes: Rather than a quote dump, the skill encodes concrete mind‑models (e.g., "social sieve", "employment back‑casting", "irreplacability test") so an agent can apply those patterns across varied user scenarios — which means responses are framework-driven and repeatable.
- Agent-ready packaging: The project follows the Agent Skills protocol and includes SKILL.md and example dialogues, making it easy for a skills-capable runtime to load the persona as a plug‑in. So what: you get a consistent decision-making engine instead of ad-hoc prompt patches.
- Research-backed voice: The skill is synthesized from books, interviews, one‑hand quotes and decision records listed in its references folder — useful when you want a persona grounded in explicit source material rather than an invented tone.
- Multi-runtime compatibility: Designed to run inside many popular skills-compatible runtimes (Claude Code, Codex, Cursor, Hermes, etc.), letting teams reuse the same persona across deployments.
Who it's for — and tradeoffs
Great fit if you are building an advice-focused agent, creating stylized educational/career guidance flows, or experimenting with persona-driven prompting and want a reproducible, research-referenced voice. Look elsewhere if you need a neutral, consensus-based assistant (the persona is deliberately opinionated and provocative), if you require multilingual first-class support beyond Chinese, or if you need quantitative evaluation benchmarks for safety and bias out of the box. The repo is a persona/skill layer — not a new model or evaluation suite — so teams should pair it with their chosen LLM and safety tooling.
