First principles thinking

Skill ysskrishna/ai-agent-skills/skills/first-principles-thinking

Structured thinking skills for AI agents: brainstorming, decision support, systems thinking, and strategy. Built for Claude Code & Cursor.

Install

npx -y skills add ysskrishna/ai-agent-skills --skill first-principles-thinking

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use this skill when the user asks for first-principles thinking or first principles (including naming them or directing use/apply/run with obvious misspellings; decisive) or wants to reason from bedrock—stripping borrowed analogies and convention, surfacing fundamentals, then rebuilding the reasoning chain and implications. Use when they want to reason from scratch, challenge industry defaults, want physics-style business breakdowns, or sanity-check whether copying incumbents still makes sense, even if they never say first principles. Skip when they want a quick convention-following checklist with no rebuild of assumptions, or purely social coordination with no modeling ask.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.2 KB, as published. Nobody here has run it

First Principles Thinking

Question inherited baggage. Rebuild only from bedrock you can defend.

How to run it with this skill: one clearly headed section per step in this order: Surface → Question → Bedrock → Rebuild → Implications.

Setup (run before starting)

In one short block:

Reconstruction target — belief, cost, design, or strategy to ground
Default pass — Surface → Question → Bedrock → Rebuild → Implications (state this line)

If immutable constraints (physics, law, budget) are unclear, ask at most 3 questions in one message, then proceed. Note unknowns or working guesses in plain language (no bracket tags in Setup).

The Steps

Surface

State the conventional answer or analogy people rely on. List loaded words or hidden comparisons ("like Uber for…").

Question

For each major assumption: Assumption: … — Why believed? (authority, analogy, experience) — What if false?

Bedrock

List fundamental truths that survive scrutiny — physics, logic identities, legal musts, documented preferences of real users, arithmetic. Label each [FUNDAMENTAL] vs [STILL ASSUMPTION].

Prefer three or more honest bedrock items when that is credible; if fewer are honest, say why in one line.

Rebuild

From only [FUNDAMENTAL] items, derive conclusions in numbered steps. No smuggled analogies; if you need a new premise, add it to Bedrock first with a tag.

Implications

So what for decisions: what changes vs the conventional path? Cost of being wrong if a tagged assumption fails.

Add a short vs convention contrast (a few bullets or a two-column mini-summary) if it clarifies the decision.

Execution Rules

Rebuild cannot cite "industry standard" as a premise unless translated into a fundamental (e.g. "buyers require SLA X because regulation Y").
If bedrock is too thin to rebuild, say insufficient grounding and list what evidence would fix it.
Avoid faux profundity; keep steps short and checkable.

Checklist (verify before responding)

Setup: reconstruction target + default pass
Surface names convention or analogy explicitly
Question step ties assumptions to why held
Bedrock uses [FUNDAMENTAL] / [STILL ASSUMPTION]
Rebuild chain only uses fundamentals
Implications name what changes vs convention (and vs-contrast if helpful)