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

feynman-perspective

Richard Feynman's mental models and communication style. Based on in-depth research across 40+ primary sources, distilling 5 core mental models, 8 decision-making heuristics, and complete communication DNA. Purpose: Serves as a thinking advisor to analyze problems, examine decisions, and provide feedback through Feynman's perspective. Trigger when the user mentions "from Feynman's perspective", "how would Feynman view this", "Feynman mode", "feynman perspective", or "Feynman technique". Also triggers if the user says "is this a cargo cult?", "naming is not understanding", "can you demonstrate instead of argue?", or "do I really understand this or did I just memorize the name?". Do not trigger on generic requests like "explain this to me" or "in simple terms"—activate only when Feynman-style validation (cargo cult detection, naming vs. understanding, anti-self-deception) is involved.

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

A pure prompt-based roleplay skill: directory contains only SKILL.md + 5 research files, with no scripts; once activated, the model answers in Feynman's first-person voice.

examples/feynman-perspective/SKILL.md
**此Skill激活后，直接以费曼的身份回应。**
What is happening here: All capabilities stem from prompts—frontmatter triggers + body text constraining persona, validation methodology, and output format.

A one-time disclaimer + explicit exit anchor to prevent persona drift.

examples/feynman-perspective/SKILL.md
🛑 **STOP（仅一次）**：首次激活时输出免责声明一次——「我以费曼视角和你聊，基于公开言论推断，非本人观点」。后续对话**绝不**重复。
What is happening here: The disclaimer appears only once; if the user says "exit/switch back to normal", it immediately reverts to standard mode.

Trigger phrases focus on "Feynman-style validation" rather than generic "explain this to me": cargo cult detection, naming vs. understanding, anti-self-deception.

examples/feynman-perspective/SKILL.md
当用户提到「用费曼的视角」「费曼会怎么看」「费曼模式」「feynman perspective」「费曼学习法」时使用。
What is happening here: The description explicitly excludes generic explanation requests and only activates when Feynman-style validation is involved.

A three-step Agentic Protocol: problem classification → Feynman-style research (first principles / experimental data / analogy boundaries / blind spots) → answering starting from concrete examples, with two CHECKPOINT gates.

examples/feynman-perspective/SKILL.md
**核心原则：费曼不猜测，他验证。他在下结论前，会先搞清楚事实是什么。这个Skill也必须这样。**
What is happening here: The research checklist emphasizes "stripping away names to see the essence" and "cargo cult detection", directly mapping to core principles.

The research step is a hard rule that requires the host environment to support web search.

examples/feynman-perspective/SKILL.md
**⚠️ 必须使用工具（WebSearch等）获取真实信息，不可跳过。**
What is happening here: The skill itself does not make network requests; real-world information retrieval is delegated to the host's search tools.

Five mental models: Naming ≠ Understanding, Anti-Self-Deception Principle, Uncertainty is Strength, Concrete Thinking, Deep Play.

examples/feynman-perspective/SKILL.md
### 模型1: 命名 ≠ 理解
What is happening here: Each model details "source evidence / application method / test questions / limitations" to avoid deifying the Feynman methodology.

Eight decision heuristics: Cargo cult detection, Demonstration > Argumentation, Reality takes precedence over narrative, Shutting down options once and for all, Concrete to general, 12-question filter, Direct verification, Anti-identity lock-in.

examples/feynman-perspective/SKILL.md
**规则**：如果一个做法有科学/专业的所有外在形式，但缺少核心精神，那就是货物崇拜——飞机不会降落。
What is happening here: Each heuristic provides application scenarios and examples (e.g., the O-ring ice water experiment, leaving Caltech comparisons).

Communication DNA (short sentence anchor followed by long sentence elaboration, conversational tone, rhetorical questions instead of exclamations, self-deprecation to build credibility) + Chinese adaptation, Failure Mode Fallbacks, and an anti-pattern blacklist.

examples/feynman-perspective/references/费曼表达风格调研.md
**短句为王，长句为辅**
What is happening here: Chinese adaptation requires using colloquial terms like "figure out / play / guess / wrong" instead of "understand / research / hypothesize / inaccurate", concluding with "that's all there is to it".

A 9-rule Failure Mode Fallback tree and an 8-rule anti-pattern blacklist to restrict jargon accumulation, pretending certainty, and going over 4 paragraphs without concrete examples.

examples/feynman-perspective/SKILL.md
异常先识别再处理；绝不静默跳过、绝不装作了解没了解过的事、绝不在身份争辩里耗时间。
What is happening here: Specifically addresses the historical constraint that Feynman passed away in 1988 and cannot hold genuine stances on AI or the modern internet.

Six honesty boundaries, the first stating: cognitive methodology is extracted without defending Feynman's personal conduct (gender controversies, etc.).

examples/feynman-perspective/SKILL.md
1. **性别问题**：费曼在对待女性方面有明确记录的问题行为（自传中物化女性的描述、FBI档案中的家暴指控）。此Skill提取的是费曼的认知方法论，不为其个人行为辩护
What is happening here: Also outlines boundaries regarding self-mythologizing, disciplinary bias, calculator vs. thinker distinctions, and historical limitations.

2Core Capabilities

01Thinking advisor mode answering in Feynman's first-person voice: activated via triggers, one-time disclaimer, explicit exit anchor.
02Agentic Protocol: problem classification + Feynman research checklist + two CHECKPOINT gates, enforcing verification before drawing conclusions.
03Five mental models (Naming ≠ Understanding, Anti-Self-Deception, Uncertainty is Strength, Concrete Thinking, Deep Play).
04Eight decision heuristics, including cargo cult detection, Demonstration > Argumentation, reality over narrative, and 12-question filter.
05Communication DNA: short sentence anchor with long sentence expansion, conversational tone, rhetorical questions over exclamations, self-deprecation for credibility, plus Chinese output adaptation.
06Failure Mode Fallback tree and anti-pattern blacklist, specifically preventing jargon accumulation and feigned certainty.
07Six honesty boundaries: gender controversies, self-mythologizing, disciplinary bias, calculator vs. thinker, historical figure limitations, unpredictability.
08Raw materials on external evaluations and controversies (Dyson/Gell-Mann/Schwinger/gender controversies) used for self-calibration rather than deification.

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 instructions and rule text, without script execution or external impacts.
Audit Notes: Grading is evaluated solely on the skill directory itself: entirely composed of pure Markdown instructions (SKILL.md + FIDELITY.md + 6 references files); 0 hits on external program execution, disk writing, network requests, dangerous flags, or sandboxing during risk scanning; 0 credential hits; the 75 URLs are purely reference link text. Therefore graded 🟢 green (pure instructions). Usage caveats (does not alter rating): web search results cannot be inherently trusted; persona-based output may amplify assertions; Feynman passed away in 1988, so the skill cannot state genuine personal stances on contemporary topics.

5Second-Pass Verification

  • [ok] 全部 evidence.quote 逐字回源 — 每条 quote 均取自 bundle 内已展开文件（SKILL.md、外部评价调研、著作与系统思考调研、表达风格调研等）的可见原文，未引用截断段落不可见文字，长度均超 12 字符下限。
  • [ok] 目录清单与 assets 一致性 — §1 列举 8 个文件（FIDELITY.md、SKILL.md、references/ 下 6 个 md），internal_assets 逐条对应，无多列少列。
  • [ok] frontmatter description 与 meta.official_desc 一致性 — 骨架 official_desc 即 SKILL.md frontmatter 描述逐字内容（含全部触发短语），保留未改。
  • [ok] 未披露脚本/网络/凭证行为 — 风险扫描显示除 URL 文本外各类均 0 命中；确认 skill 目录内无可执行代码，WebSearch 仅出现在提示词层。
  • [discrepancy] research.md 内部路径可定位性 — research.md 的「详细调研报告位置」表写成 `07-调研与分析/费曼著作与系统思考调研-20260404.md` 等 skill 目录外路径；实际文件位于 examples/feynman-perspective/references/ 下，索引与实际布局不符。
  • [unlocatable] FIDELITY.md 与 费曼重大决策调研 正文 — §1 列出 examples/feynman-perspective/FIDELITY.md（2729 bytes／20 行）与 references/费曼重大决策调研-20260404.md（18000 bytes／239 行），但 bundle §2 未展开其正文，故本次不对其内容作任何声称，仅记录存在与大小。
  • [ok] commit / 授权 / 调研时间 — skill.commit 与 audit.skill_commit 同为 fe0374687037c4cc51a65c1e0c145afe2981dc69；授权为仓库根 MIT；调研截止 2026-04-04，与文档「调研时间：2026-04-04」相符。

6Verdict & Scenarios

  • Enforces 'verification first' as an unskippable workflow: Step 1 mandatory classification, Step 2 mandatory evidence gathering, two CHECKPOINT gates, with 'cargo cult detection' built directly into the research checklist.
  • Maintains a clear 'non-deification' discipline: honesty boundaries directly address personal controversies and domain biases, with external evaluation files compiling sharp critiques.
  • Core concepts are traced back to original sources, distinguishing primary material from later distillations.
  • Explicitly prohibits flawed patterns during training/usage.
  • Chinese adaptation requires natural colloquial phrasing rather than generic AI tone.
    • Best for: Best for: Anyone wanting to test if they "truly understand" a concept, identify cargo-cult processes, explain complex issues via analogies and concrete experiments, or push forward amidst uncertainty; education, science communication, and technical writing scenarios; users needing "anti-self-deception" self-calibration. Also serves as a reference model for handling historical limitations and personal controversies in persona skills.
    Not for: Not suitable for: Scenarios requiring delicate social diplomacy or team emotional management; questions requiring an impartial evaluation of humanities or philosophy; users seeking genuine stances on modern (post-1988) topics; lightweight Q&A where users only want a quick explanation without going through a "verify first, answer later" process.
    Install Agent-Ready
    ① Verified Mirror 2026-10-10
    Option A · Direct DownloadDownload feynman-perspective.tar.gz
    sha256: 21ca4c809de3918f…
    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.