# elon-musk-skill is a prompt file that distils a persona, and asks you to ship tomorrow > elon-musk-skill is an Agent Skills package that installs a writing and reasoning framework for Elon Musk into any compatible agent runtime, built from four research files and 30 plus primary sources. It is a persona prompt, and it is explicit about the trade-offs in doing that. **alchaincyf/elon-musk-skill** — 马斯克.skill — Elon Musk的认知操作系统。由女娲.skill生成。 - Repository: https://github.com/alchaincyf/elon-musk-skill - Stars: 551 · Forks: 118 - Language: Unknown - License: MIT - Published: 2026-10-22 · Updated: 2026-10-22 · Language: en - Canonical page: https://hysenlabs.com/projects/alchaincyf-elon-musk-skill ## A skill, not a library, and not a quote collection elon-musk-skill is presented as an operating system for one person's cognition, and the README makes a point of what it is not: not a collection of quotes, but a runnable thinking framework. It is built on the open Agent Skills protocol, which means it is a directory of files that an agent runtime loads, not a package with an API. The README lists the compatible runtimes as Claude Code, Codex, Cursor, OpenClaw, Hermes Agent, CodeBuddy, Workbuddy, Gemini CLI and OpenCode, and claims 50 plus runtimes, with a badge for multi-runtime support. The substance of the skill, according to the README, is 5 core mental models, 8 decision heuristics and a complete expression DNA, extracted from 30 plus primary sources through deep research: the Walter Isaacson biography, long podcast conversations with Joe Rogan and Lex Fridman, an Everyday Astronaut factory visit, continued output on X and Twitter, court testimony and SEC filings. The claim the README makes is specific: each response applies a concrete mental model, and it does not recite quotes, it uses the framework to break down your question. The named models are the asymptotic limit method, the five step algorithm, existential anchoring, vertical integration as physical necessity, and fast iteration over perfect planning. The MIT licence, the primary language listed as unknown, and a repository that is not archived with the last push on 2026-10-13 complete the picture. It is a prompt artefact in a markdown file, and the repository is small: README.md, SKILL.md, a LICENSE, assets/, examples/, references/ and a wechat-qrcode.jpg. ## Installing it, and the two ways to run it The recommended installation is a single command, and the installer in question is the general purpose CLI from vercel-labs/skills, which supports 55 plus runtimes. ```bash npx skills add alchaincyf/elon-musk-skill ``` The README says this installer detects the current runtime and puts the skill in the right directory, and that you can pass a flag such as -a claude-code, -a codex, -a cursor or -a openclaw when you need to name the runtime explicitly. The manual route is a clone into the runtime's own skills directory, with a runtime flag or argument to select where it lands. The README tabulates those paths, and they are conventional: ~/.claude/skills/elon-musk-skill/ for Claude Code, ~/.codex/skills/elon-musk-skill/ for Codex CLI, ~/.cursor/skills/elon-musk-skill/ for Cursor, and ~/.openclaw/workspace/skills/elon-musk-skill/ for OpenClaw, with Hermes Agent handled by running that runtime's install script or cloning into its skills directory. There is a third option that is worth noticing because it removes the runtime requirement entirely. Even if a runtime does not support automatic Agent Skills loading, you can paste the contents of SKILL.md into the conversation, because it is essentially a markdown file with YAML frontmatter. That is the honest description of the whole thing. The skill is a prompt. The runtime support is a convenience for loading it automatically, and the value of the package is the content of SKILL.md and the research behind it, not the packaging. ## The five mental models, and the heuristics built on them The README tabulates the 5 mental models with a one line summary and a source, and reading the table tells you what the skill is actually teaching an agent. The asymptotic limit method is listed as: calculate the theoretical optimum the laws of physics allow, then ask why reality is so far from it, sourced to the SpaceX rocket cost breakdown and the Tesla battery cost analysis. The five step algorithm is questioning the requirement, deleting, simplifying, accelerating and automating, with the order not to be rearranged, sourced to the Everyday Astronaut factory visit where it was first stated in full. Existential anchoring ties every decision to the scale of human civilisation continuing, turning small failures into acceptable costs. Vertical integration as physical necessity is argued from a high idiot index, meaning the supply chain middle layer collects a tax on information opacity. And fast iteration over perfect planning treats an aggressive timeline as a management tool and accepts many failures as the price of learning faster. The 8 decision heuristics are shorter and more operational. Every requirement carries a name, and always having done it this way is not accepted. Calculate the asymptotic limit first, and a gap above 5 times means huge room for improvement. Delete past the point of comfort and add back, and if 10 percent did not come back you did not delete enough. Making is harder than designing by a factor of 10, so do not linger on paper. The laws of physics are the only hard constraint, and rules and conventions can be challenged. Get in the room and solve the most critical bottleneck yourself. Use resources across companies as a lever, such as your own rocket launching your own satellite. And an aggressive timeline is a pressure tool, trading reputation for speed. The idiot index, in the vocabulary list, is a recurring invented term: what you pay a cloud provider divided by the actual cost of the compute underneath. ## Five documented tensions, kept on purpose The most interesting part of the README is a section titled the 5 internal tensions, and it argues that this is not a caricature of the engineering maniac. The skill keeps the contradictions: an AI worrier who is also an AI developer, warning about the threat while starting xAI. Free speech alongside banning critics, claiming absolutism while blocking the account that tracks his plane. A rational framework alongside emotional outbursts, with the five step algorithm described as extremely rational and a demon mode that roars at executives. Aggressive transparency alongside selective silence, saying what he thinks while being strategically absent from court. Failure as innovation alongside intolerance of dissent, encouraging engineering failure while dismissing an employee who expressed disagreement. Then there is the expression DNA section, which is a writing style spec rather than a set of beliefs. Sentence form is a minimalist manifesto, short phrases of three to six words, carved like an inscription rather than written like an email. Rhythm is conclusion first and reasoning after, improvised cost breakdowns, an apology that cuts into an attack without a seam. Vocabulary is engineering terms made ordinary, with asymptotic limit, idiot index and first principles. Humour is status deflation, a billionaire posting a meme, and provocation, turning the SEC into entertainment, deliberately cringe. Attitude is opposition rather than compromise, probabilistic self-description, and a refusal to answer inside someone else's frame. Whether an agent reproduces any of that faithfully is an empirical question the repository does not answer, and the README does not claim measurements. ## How the skill was built, and the generator behind it The README explains the provenance, and the process is the more transferable idea in the repository. The skill was generated by another skill, nuwa-skill, and the badge says it was made with that tool. The described workflow is: input a name, six agents research in parallel across books, conversations, expression, criticism, decision making and timeline, the results are cross-validated and distilled into mental models, a SKILL.md is built, and then quality verification runs with three known tests, one edge case test and a style test. So the pipeline is parallel research, cross validation, synthesis, then a test harness. The tests are the interesting part, because a persona prompt is exactly the kind of artefact where a plausible-sounding failure is invisible without a check. The generator is installable in the same way. ```bash npx skills add alchaincyf/nuwa-skill ``` And the README invites you to distil other people the same way, by asking for it. There is a table of other figures already done: Steve Jobs for product, design and strategy, Naval for wealth, compounding and life philosophy, Charlie Munger for investing, multi-discipline thinking and inversion, and Richard Feynman in the same table, with more cut off in the visible text. The licence is MIT, and the README notes the animation in the header was made by the huashu-design skill. There is also a wechat-qrcode.jpg at the repository root, which suggests the project has an audience that is not reached through GitHub alone. ## The research files, and what is deliberately left out Four research files sit in the references/ directory, and the README tabulates them. research.md is the combined research covering biography distillation, thinking models and expression style. Two files dated 20260404 are a systematic breakdown of the thinking system and an analysis of decision patterns and behaviour. A fourth covers improvisational thinking and expression. The primary sources are named: the Walter Isaacson biography from 2023, Ashlee Vance's biography, the X and Twitter account, multiple episodes of the Joe Rogan Experience, multiple Lex Fridman episodes, a 2022 TED talk, the Everyday Astronaut factory visit, the All-In Podcast, court testimony and SEC filings, and SpaceX and Tesla earnings calls. The external critical sources are listed separately, and they are not gentle: DOGE layoff effectiveness assessments, FSD timeline commitment tracking, analysis of the Twitter and X acquisition aftermath, former employee accounts, and SEC litigation records. One sourcing decision stands out. The README states that information sources have excluded Zhihu, WeChat public accounts and Baidu Baike. That is a deliberate restriction on what counts as research input, and it says something about the target audience and about the author's view of source quality. The repository also has an examples/ directory with a single file, demo-conversation.md, described as a complete four round record of a real conversation. The README shows excerpts from it, in the format of a two column exchange with a character marker for the persona on one side. That is a useful thing to have in the repository, because it shows the intended output shape rather than describing it. ## Where elon-musk-skill is the wrong choice Three cases, and the first is a matter of what you are actually asking for. This is a persona prompt. It makes an agent argue in a particular voice, apply a particular set of heuristics and reach for particular framings, the asymptotic limit and the five step algorithm among them. It does not contain a model of how a company should be run, and it does not have evidence that its advice improves decisions. If you want a business framework, use a business framework. If you want an agent that challenges your assumptions in an abrasive register and pushes you toward shipping, that is what this is, and the README is careful to say it is a thinking framework rather than a quote collection rather than anything predictive. The second is the register itself. The heuristics include accepting reputational damage in exchange for speed, treating failure as learning, and a method of removing requirements aggressively, deleting past the point of comfort and adding back 10 percent. Applied to a consumer product, that is a reasonable technique. Applied to safety critical software, medical devices, financial systems or anything with regulatory consequences, an agent instructed to remove requirements and ship tomorrow is a bad idea, and the skill offers no such guardrail. The third is maintenance. This is a prompt in a markdown file, so it has no tests in the sense a library has, it does not version its content, and the research files carry a fixed date. The generator that produced it is the more durable artefact, because a method for building persona skills survives its subjects better than any single persona does. One practical point to check before installing. Because the same installer places the skill into per-runtime directories, running the install command twice for two runtimes copies the skill in two places, and any later edit to one copy is not reflected in the other. If you use more than one runtime, decide which copy is the source of truth. ## Conclusion Adopt elon-musk-skill if you want a structured way to interrogate cost structures, product scope and iteration speed with an agent, and you are comfortable with a persona that argues rather than agrees. Do not adopt it as an oracle about what Musk would actually decide, since the README describes a distilled framework rather than a simulation, and treat its advice as a lens rather than a strategy. Before you install, read SKILL.md and the four files in references/ so you know what the agent is being told to imitate, and check the runtime directory it lands in so it does not collide with an existing skill. ## FAQ ### What is elon-musk-skill? It is an Agent Skills package that installs a runnable thinking framework for Elon Musk into compatible agent runtimes, built from 30 plus primary sources and distilled into 5 mental models, 8 decision heuristics and an expression style. The README states it is not a quote collection but a framework the agent applies to your question. ### How do I install elon-musk-skill? Run npx skills add alchaincyf/elon-musk-skill, which detects the current runtime and places the skill in the right directory, or clone the repository into the runtime's skills directory manually. You can also paste the contents of SKILL.md into a conversation if your runtime does not support automatic skill loading. ### Which agent runtimes support elon-musk-skill? The README lists Claude Code, Codex, Cursor, OpenClaw, Hermes Agent, CodeBuddy, Workbuddy, Gemini CLI and OpenCode among the compatible runtimes, and the installer from vercel-labs/skills is said to support 55 plus runtimes and auto-detect the current one. ### How was elon-musk-skill created? It was generated by another skill, nuwa-skill, whose workflow is described as taking a name, running six agents in parallel to research books, conversations, expression, criticism, decision making and timelines, cross validating to distill mental models, building SKILL.md, and then verifying quality with three known tests, one edge case test and a style test. ## Sources - [alchaincyf/elon-musk-skill on GitHub](https://github.com/alchaincyf/elon-musk-skill) - [Issues](https://github.com/alchaincyf/elon-musk-skill/issues) - [License: MIT](https://github.com/alchaincyf/elon-musk-skill/blob/main/LICENSE) - [README](https://github.com/alchaincyf/elon-musk-skill/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/alchaincyf-elon-musk-skill