Home / All Skills / Workflow & Core Tools / mrbeast-perspective
English (EN)简体中文 (ZH)日本語 (JA)
Workflow & Core Tools · alchaincyf/nuwa-skill

mrbeast-perspective

MrBeast's (Jimmy Donaldson) content creation operating system. Built upon in-depth research from a leaked 36-page internal training manual, 6 in-depth podcasts, decision logs, and external critiques, it extracts 6 core mental models, 8 decision heuristics, complete formulas for titles/thumbnails/hooks/pacing, and 4 runnable content analysis scripts. Once activated, it immerses you in a roleplay of MrBeast, offering content creation advice directly from a first-person "I" perspective. Trigger it when users mention "from MrBeast's perspective", "what would MrBeast do", "Beast mode", or "mrbeast perspective". It should also trigger even if the user simply asks "how to increase video CTR", "title is not catchy enough", "how to optimize the retention curve", or "should I change the thumbnail". Do not trigger on general questions like "content creation advice" or "how to make content"—only activate when video optimization and YouTube methodologies such as titles/thumbnails/hooks/retention rates are involved.

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandInstall DepsNetwork AccessWrite FileDual-pass Audit
Author alchaincyfGitHub alchaincyf/nuwa-skill ↗License MIT（仓库根 LICENSE）Commit fe03746870
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Roleplay shell: Responds directly in the first person as MrBeast (Jimmy) upon activation, delivers a disclaimer only once during the initial activation, and immediately breaks character when hit by exit keywords.

examples/mrbeast-perspective/SKILL.md
**此Skill激活后，直接以Jimmy/MrBeast的身份回应。**
What this does: Turns the "perspective" into a toggleable immersive persona—utilizing both a "one-time disclaimer + EXIT TRIGGER" constraint to maintain immersion while preserving an exit path, sharing the same skeleton with the nuwa series in the same repository.

Enforced Agentic protocol: First classifies the question type; questions involving specific channels, videos, or platform data must conduct online research first before answering with mental models.

examples/mrbeast-perspective/SKILL.md
**核心原则：我不猜，我测。在给内容建议之前，先看数据。这个Skill也必须这样。**
What this does: Codifies "data-driven" into an unskippable workflow (Step 1 Classification → Step 2 WebSearch → Step 3 Answer), keeping research summaries organized internally without outputting them to the user.

Three-stage CHECKPOINT self-check, anchoring output quality to "deliver judgments straightforwardly, ensure every piece of advice is actionable, and provide no fake encouragement".

examples/mrbeast-perspective/SKILL.md
1. **第一句是判断还是铺垫**？必须开门见山，「这个标题有三个致命问题」式开头。
What this does: Inserts mandatory self-checks at critical nodes to prevent falling back to empty platitudes like "titles should be catchy"; Checkpoints A, B, and C cover classification, research, and output respectively.

9-rule if-then-then failure mode Fallback tree, providing primary actions and fallback phrasing item by item.

examples/mrbeast-perspective/SKILL.md
| 1 | WebSearch 搜不到具体数据/Benchmark | 换 query：加「VidIQ / SocialBlade / 频道名 + 月增长」等限定词 | 直接对用户说：「告诉我 3 个具体数字——你最近 3 个视频的 CTR、AVD 和订阅净增。没有数据我不能瞎给建议」 |
What this does: Maps every deviation to a fallback action, focusing on turning "lack of data" into asking the user for 3 key metrics rather than hallucinating.

7-item anti-pattern blacklist, explicitly prohibiting the output formats most prone to faulty imitation (listicles, vague encouragement, direct cross-platform copy-pasting, etc.).

examples/mrbeast-perspective/SKILL.md
| 5 | 给 listicle 风格的建议（「10 个 YouTube 增长技巧」） | listicle 标题在 2024 已经死了，Jimmy 不会给这种建议 | 用「这个标题有 3 个致命问题」「下一个视频做这 3 个动作」式直接判断 |
What this does: Uses three columns—"bad example / why it's wrong / correct approach"—to enforce negative constraints, which suppress the model's default writing style more effectively than positive descriptions.

Built-in 6 core mental models (CTR × AVD, Zero Boring Moments, Escalating Progression, Simple Concept × Extreme Execution, Total Reinvestment Flywheel, Creative Frugality).

examples/mrbeast-perspective/SKILL.md
**一句话**：YouTube上只有两个数字重要——点击率（Click-Through Rate）和平均观看时长（Average View Duration）。其他一切都是噪音。
What this does: Equips each model with formulas, sources, original quotes, and limitations, asking "will this increase CTR or AVD?" before making decisions; Model 4's formula defines Virality = Concept Simplicity × Execution Extremity.

Includes 4 runnable scripts (subtitle downloader + title analyzer + retention checker + thumbnail auditor), translating methodology into actionable audits.

examples/mrbeast-perspective/SKILL.md
| `fetch_youtube_subtitles.sh` | 下载YouTube视频字幕 | `./fetch_youtube_subtitles.sh <URL> [lang]` |
What this does: scripts/ provides three local Python analyzers and one shell script; the shell script checks/installs yt-dlp and downloads subtitles, acting as the only entry point for network requests and package installation in this skill.

Explicit honesty boundaries: Declares YouTube ≠ all platforms, budget disparities, English ≠ Chinese market, philanthropy and employee treatment controversies, with research cutoff at 2026-04.

examples/mrbeast-perspective/SKILL.md
1. **YouTube ≠ 所有平台**。我的方法论对YouTube优化最深，B站、抖音、公众号的算法和用户行为不同，需要翻译而非照搬。
What this does: Brings the methodology's scope of application and known controversies upfront, avoiding applying a $400M-scale playbook to zero-budget creators.

2Core Capabilities

01Immersive MrBeast roleplay with an explicit exit protocol
02Complete YouTube content optimization formulas: title patterns, thumbnail 3-element rule, first 30-second hook, retention segmented pacing
03Mandatory online research protocol (selects research dimensions by question type: data/competitors/trends/cost-benefit)
04Output fallback: 9 if-then-then failure modes and fallback responses
05Style correction: 7 anti-pattern blacklists, rewriting vague talk into actionable directives
066 mental models library + 8 decision heuristics + thumbnail/title self-checks (e.g., Zoom Out Test)
074 runnable content analysis scripts (subtitle downloader / title analyzer / retention checker / thumbnail auditor)
08Honesty boundary statements (inapplicability regarding platforms/budgets/languages and known controversies)

3External Dependencies

KindDependency / Endpoint
cliyt-dlp（fetch_youtube_subtitles.sh 调用；缺失时自动 pip 安装）
api宿主 WebSearch / 联网研究（正文强制要求）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Graded/per-use authorization, sandbox recommended
For Administrators
Approval workflow, rollback support, full audit
Evidence & Triggers (File Paths & Excerpts)
  • scripts/: Executes local scripts or commands: examples/mrbeast-perspective/scripts/fetch_youtube_subtitles.sh (invokes yt-dlp to download subtitles)
  • scripts/ 或 SKILL.md: Declares dynamic installation of external language package dependencies
  • 网络端点: Declares external network access: examples/mrbeast-perspective/scripts/fetch_youtube_subtitles.sh accesses YouTube via yt-dlp
  • 本地工作目录: File read/write and artifact output: examples/mrbeast-perspective/scripts/analyze_titles.py writes reports to --output: Path(
Audit Notes: The security rating is based solely on the examples/mrbeast-perspective/ directory itself: scripts/fetch_youtube_subtitles.sh automatically runs pip install -q yt-dlp when yt-dlp is missing (relying on remote packages, supply chain surface), and sends user-provided URLs to YouTube via yt-dlp; the other three Python scripts only perform local text analysis and write reports to --output. Within the directory, there is no credential/cookie/env reading, no protection bypass, no security downgrade, and no arbitrary code execution surface. SKILL.md requires the agent to use the host's WebSearch, which constitutes host behavior and is not factored into this rating.

5Second-Pass Verification

  • [ok] 宣称的 6 心智模型 / 8 决策启发式是否在 SKILL.md 内可落地 — SKILL.md 逐个给出「一句话 + 原理/来源 + 应用 + 局限」，模型 1/2/4 还附原话引用，不是只列表名。
  • [ok] 4 个脚本是否真的存在于本 skill 目录 — §1 清单含 scripts/ 四个文件：analyze_titles.py 11004B、fetch_youtube_subtitles.sh 3150B、retention_curve_checker.py 15396B、thumbnail_audit.py 14560B。
  • [ok] 联网与外发风险定位到哪一段 — §3 显示 fetch_youtube_subtitles.sh 第 38-41 行检查/安装 yt-dlp（pip install -q yt-dlp），第 50 行 mkdir -p "$OUTDIR"；这是本 skill 唯一的联网与远程包入口，据此判 🟠。
  • [ok] 负面信息是否被隐瞒或美化 — §2 的 04-external-views 与 SKILL.md 诚实边界都主动写入 Beast Burger、Lunchly、员工待遇、Beast Games 诉讼与「poverty porn」批评，未做单向美化。
  • [ok] 调研时效声明与站内日期是否自洽 — SKILL.md 诚实边界写「调研截止2026年4月」，§2 时间线覆盖到 2026-04；与站内分析日期 2026-10-10 之间约有半年空档，属文档已声明的边界。
  • [ok] 是否会实际执行脚本 — SKILL.md 只在工具表给出用法（如 python analyze_titles.py titles.txt），未见任何自动调用；是否执行取决于用户/agent。
  • [ok] meta.installs 是否可定位单一 skill — 无 per-skill 安装量可定位，installs 保持 null。
  • [ok] 脚本写盘范围 — 三个 .py 的 write_text 目标由 --output 参数决定，属本地写入；shell 脚本写往 $OUTDIR，均非系统路径。

6Verdict & Scenarios

  • Translates creator methodology into actionable checks: includes title/retention/thumbnail analyzers instead of merely preaching principles
  • Extremely strong output style constraints: CHECKPOINT × 3 + Fallback × 9 + Anti-pattern Blacklist × 7, significantly suppressing empty platitudes and insincere praise
  • Does not shy away from negative facts and honesty boundaries (Beast Burger issues, employee burnout, and philanthropic ethical critiques are written into the main text)
  • Full-chain methodology coverage: from mental models and decision heuristics to title/thumbnail/hook/pacing formulas, ready for direct execution
    • Best for: Best suited for: (1) YouTube creators needing to optimize titles, thumbnails, the first 30 seconds, or retention curves; (2) Content teams looking to turn "data-driven + extreme execution" into an actionable checklist; (3) Scenarios exploring content topics and growth strategies from MrBeast's perspective; (4) Use cases requiring local scripts to quickly audit batches of titles or script retention.
    Not for: Not suited for: (1) Creators focusing on Bilibili/Douyin/WeChat Official Accounts who are unwilling to adapt to platform differences (algorithms and user behaviors differ); (2) Those needing budget-neutral, generic content advice (this skill defaults to a high-budget context); (3) Anyone seeking gentle encouragement or a corporate PR tone; (4) Scenarios that do not tolerate first-person roleplay.
    Install Agent-Ready
    ① Verified Mirror 2026-10-10
    Option A · Direct DownloadDownload mrbeast-perspective.tar.gz
    sha256: 5b5c06118d80624d…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit fe03746870.