Hubbestaiskill.com

Home / All Skills / Workflow & Core Tools / zhang-yiming-perspective

Workflow & Core Tools · alchaincyf/nuwa-skill

zhang-yiming-perspective

Zhang Yiming's (Founder of ByteDance/TikTok) thinking framework and communication style. Based on research across 6 dimensions (writings, in-depth interviews, expressive DNA, third-party perspectives, decision logs, and timelines), covering 32 interview excerpts and 12 major decision cases, distilling 5 core mental models, 7 decision heuristics, and a complete expressive DNA. Purpose: Serves as a strategic thinking advisor analyzing products, organizations, globalization, talent, and personal growth through Zhang Yiming's lens. Trigger when the user mentions "from Zhang Yiming's perspective", "how would Zhang Yiming see this", "Yiming's approach", or "zhang yiming perspective". It should trigger even if the user simply says "help me think about this from Zhang Yiming's angle", "what would ByteDance do", or "switch to Zhang Yiming". It should also trigger on "ByteDance's view", "Toutiao's logic", "how would Yiming choose", or "Yiming".

L1 Basic · Source registration onlyV1 Commit-PinnedDual-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 as the first principle: Once activated, respond directly using the first-person "I" as Zhang Yiming, adopting his tone and pacing; state disclaimers only once upon initial activation, avoiding meta-analyses or phrases like "If Zhang Yiming were to...".

examples/zhang-yiming-perspective/SKILL.md

**此Skill激活后，直接以张一鸣的身份回应。**

What this does: Enforces "roleplay" as a hard constraint to prevent the model from breaking character to deliver third-party commentary.

Anti-drift design: EXIT TRIGGER allows exiting character at any time, actively steps down complexity during long-dialogue wrap-ups, and prohibits a preachy tone via "role drift warnings".

examples/zhang-yiming-perspective/SKILL.md

如果输出开始出现「我认为大家应该……」「社会需要……」等说教语气，立即停止

What this does: Provides exit hatches and self-monitoring checkpoints for roleplay to prevent persona breakdown or dogmatic tones.

CHECKPOINT three questions running across Step1→2→3: Self-verify before output that the opening sentence must be a verdict (not introductory context), "I noticed" appears at most twice, and assess how many mental models were applied (≤2 and do not cite model indices).

examples/zhang-yiming-perspective/SKILL.md

第一句是判断还是铺垫？必须是判断，不要先讲背景。

What this does: Converts communication style guidelines into hard checkpoints to ensure a "conclusion-first" linguistic rhythm.

Three-step Agentic workflow: Step 1 classifies queries into "fact-dependent / pure framework / hybrid", Step 2 must conduct Zhang Yiming-style research via tools like WebSearch, and Step 3 delivers verdicts strictly grounded in factual data.

examples/zhang-yiming-perspective/SKILL.md

**⚠️ 必须使用工具（WebSearch等）获取真实信息，不可跳过。**

What this does: Constrains the model through a "verify facts before forming verdicts" approach to prevent hallucinated answers about products or events based solely on pre-training data.

Step 2 research is structured across four dimensions: information distribution efficiency, organization, globalization, and data flywheels; research findings are internally compiled into a factual brief (not outputted), surfacing only fact-based assessments to the user.

examples/zhang-yiming-perspective/SKILL.md

1. **这个产品/系统的信息分发效率如何**

What this does: Pre-configures analysis lenses for "what to look up", preventing cluttered search dumps while hiding the raw investigative process from the user.

A 9-rule Failure Mode Fallback tree assigns remediation actions based on signals—e.g., forcing WebSearch when post-2024 events are involved but Step 2 was bypassed, or prioritizing facts and projecting them into underlying layers when facts contradict prior stances.

examples/zhang-yiming-perspective/SKILL.md

| 2 | 涉及 2024 年后事件但跳过 Step 2 | 强制 WebSearch | 「我等一下，我不靠记忆判断这件事。」 |

What this does: Predefines recovery routes for common failure modes such as "answering blindly from memory" or "reciting model numbers".

7 anti-pattern blacklists constraining expressive DNA: bans emotional rallying phrases, forbids citing model numbers or exposing analytical workflows, caps "I noticed" frequency, prohibits quoting investment figures like Charlie Munger, and bans moralistic declarations and rigid narrative arcs.

examples/zhang-yiming-perspective/SKILL.md

| 5 | 引用芒格 / 塔勒布 / 巴菲特等投资圈人物 | 不是张的引用谱系 | 引乔布斯 / 稻盛和夫 / 工程师文化 / 推荐系统术语 |

What this does: Establishes clear tone boundaries via a blacklist to prevent the persona from dissolving into generic AI phrasing.

Core knowledge consists of 5 mental models (Delayed gratification is cognitive boundary, Project surface issues into high-dimensional simple problems, Algorithms are tools while empathy is the root, Negative economies of scale and Context not Control, Escaping the gravity of mediocrity) plus 7 decision heuristics, accompanied by a FIDELITY.md fidelity scorecard.

examples/zhang-yiming-perspective/SKILL.md

**一句话**：所有复杂问题都是底层简单问题的投影。不要在表象层优化，要往底层挖。

What this does: Distills the subject's worldview into reusable analytical lenses, providing stable frameworks rather than generic conversational outputs.

2Core Capabilities

01Zhang Yiming persona roleplay (first-person, concise conclusion-first delivery, low emotional tone)

02Fact calibration and four-dimensional research via WebSearch (information distribution efficiency / organization / globalization / data flywheel)

03Projecting surface symptoms into underlying high-dimensional simple problems (Mental Model ②)

04Organizational and information systems diagnostics (negative economies of scale, upward management, Context not Control)

05Talent assessment and "overfitting" judgments (hiring philosophy, empathy as foundation, independent thinking and practice)

06Silence and pivot tactics on political/regulatory topics (no stances taken, reframing into analyzable operational dimensions)

07Long-term perspective and career anxiety mitigation (delayed gratification, escaping mediocrity, gathering biographical samples)

08Anti-drift and stylistic self-checks (EXIT TRIGGER, CHECKPOINT, anti-pattern blacklists)

3External Dependencies

Kind	Dependency / Endpoint
cli	宿主 WebSearch / 联网检索工具（Step2 张一鸣式研究必需；由宿主提供，非 skill 自带脚本）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Can be auto-loaded

For Administrators

Register source and version

Evidence & Triggers (File Paths & Excerpts)

SKILL.md: Pure prompt directives and rule text, without script execution or external side effects

Audit Notes: Grading basis (evaluating only this skill directory): examples/zhang-yiming-perspective consists entirely of Markdown files (SKILL.md, FIDELITY.md, 6 research documents under references/research/), containing no scripts, binaries, endpoints, credential/environment variable accesses, or disk-write operations, qualifying as "pure instructions" tier and graded green. Non-directory aspect clarification (does not alter rating tier): Step 2 relies on the host's WebSearch/network tools, which are host capabilities rather than local code; ingesting this external content introduces a prompt-level injection surface (see injection_surface). No dangerous flags, protection bypasses, or arbitrary code execution.

5Second-Pass Verification

[ok] 角色扮演规则与 EXIT TRIGGER 属实 — SKILL.md「角色扮演规则」段确有「直接以张一鸣的身份回应」「免责声明仅首次激活时说一次」「不跳出角色做meta分析」；EXIT TRIGGER 段列出「退出」「切回正常」「不用扮演了」「stop」「停一下」五类触发词，与 claim 一致。
[ok] Agentic 三步与 Step2 四维度属实 — Step1 分类表（需要事实/纯框架/混合）与 Step2 的四个小节（看信息效率/看组织/看全球化/看数据飞轮）逐条可在内文找到，且含「必须使用工具（WebSearch等），不可跳过」的强制句。
[ok] CHECKPOINT 三问与反机械化约束一致 — Step1→2、Step2→3、Step3 输出前三组自查问题均存在；表达 DNA 段的「反机械化约束」列出「我发现最多 2 次」「不确定性收尾不是每次必须有」「叙事弧线要变化」「工具调用不可见」，与 CHECKPOINT 口径吻合。
[ok] 5 模型 + 7 启发式与 FIDELITY 描述互证 — 正文核心心智模型确为 ①延迟满足 ②投影高维 ③算法工具/同理心 ④负规模效应与 Context not Control ⑤逃逸平庸；决策启发式确为 7 条。FIDELITY.md 亦写「心智模型5个（各含证据+应用+局限）」「失败模式9条」「反例黑名单7条」，两处计数一致。
[ok] §3 风险扫描 69 条 URL 的性质归类 — 抽样命中（sourcecodecap.com 原文链接、jiemian/gamelook/wallstreetcn/guancha/x.com/thepaper、36kr、pingwest、interconnected.blog 等）全部位于「来源清单/调研来源」与研究文件里，属文档化引用；无运行时 API 端点、无 webhook；凭证类与执行/写盘/网络请求类均 0 命中。
[ok] 目录内无脚本 / 无凭证读取 / 无写盘 — §1 全部 8 个文件类型均为 text，§3「执行外部程序/写盘/网络请求/危险旗标/沙箱」五类均 0 命中；SKILL.md 仅以自然语言要求宿主使用 WebSearch，无本目录代码可执行。

6Verdict & Scenarios

High persona fidelity: Hardcodes Zhang Yiming's linguistic style (short sentences leading with conclusions, mathematical/probabilistic framing of emotional topics, direct English word integration, low affect, no coddling) and mental models, verified by an independent dual-agent scored FIDELITY.md (93/100, full score on 3 stance benchmarks).

Fact-first Agentic design: Mandatory WebSearch + a 4-dimensional research framework (info efficiency, organization, globalization, data flywheel), explicitly prioritizing "search one more time rather than hallucinating from pre-trained weights", paired with 9 fallback remediation paths.

Comprehensive anti-drift engineering: EXIT TRIGGER, 3-question CHECKPOINT, 7 anti-pattern blacklist rules, 9 failure modes, role-drift alerts, anti-mechanization constraints, and variable narrative arc rules to prevent persona breakdown.

Honest boundaries and documented tensions: Explicitly documents 7 honesty boundaries + 4 unresolved internal tensions (algorithmic neutrality vs. platform responsibility, delayed gratification vs. Douyin's instant gratification, Context not Control vs. high centralization, domestic compliance vs. international refusal to compromise), maintaining rare intellectual honesty.

Best for: Ideal for users seeking to review product, organizational, globalization, talent, and personal growth decisions through Zhang Yiming's lens (rational, delayed gratification, projecting surface problems into underlying structures); also serves as an exemplary template for "Persona Perspective + Agentic" skills—fully equipped with role rules, CHECKPOINTs, anti-pattern blacklists, failure mode trees, and fidelity scorecards.

Not for: Not suitable for scenarios requiring authentic, citable verbatim quotes from Zhang Yiming himself or unreleased personal stances (outputs are simulated; citing them constitutes misattribution); not suitable for substantive analyses of sensitive political/regulatory issues (the skill explicitly declines taking stances, redirecting solely to analyzable operational dimensions); not suitable for offline environments lacking WebSearch capabilities—the absence of Step 2 significantly impairs the factual accuracy of product and event-related responses.

Install Agent-Ready

① Verified Mirror 2026-10-10

Option A · Direct DownloadDownload zhang-yiming-perspective.tar.gz

sha256: e5392e2a41d89795…

Option B · Copy Guide for Agent

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit fe03746870.