[agents](https://carteblanche.com/index.md?kind=agent) / [research](https://carteblanche.com/index.md?category=research) # first-principles-thinking what it does Use when the user wants to challenge assumptions, break down a complex problem from scratch, or approach something with first principles reasoning. Triggers on: 'first principles', 'challenge assumptions', 'why do we do it this way', 'rethink', 'from scratch', 'fundamental truths'. tools it asks for Reads files · Uses the web by VoltAgent · version 1 · license MIT We imported this agent from its author’s repository. Is this yours? [Claim this listing](https://carteblanche.com/claim/first-principles-thinking) Verified · Before and after: 2 models · 1 try per case · [View source](https://github.com/VoltAgent/awesome-claude-code-subagents/blob/ed74d115b7fb7026a22a4f816a8e3250f425e3b4/categories/10-research-analysis/first-principles-thinking.md) · [Install from your agent](https://carteblanche.com/connect.md) · [Download .zip](https://carteblanche.com/skills/first-principles-thinking/download) ## Compare the answers Answers from Claude Sonnet 5.5 and GPT-6.1 Sol · try 1 of 1 Same request, same model, once with the agent’s instructions as its system prompt and once without. Neither could use a tool. Read both and judge for yourself. [Compare the answers](https://carteblanche.com/skills/first-principles-thinking/compare) ## The models 2 models Claude Sonnet 5.5 made by Anthropic claude-sonnet-5-5 · revision not reported by the provider 1 try per case · tested 7 Nov 2026 GPT-6.1 Sol made by OpenAI gpt-6.1-sol · revision not reported by the provider 1 try per case · tested 7 Nov 2026 ## The tasks written by Carte Blanche **example-1** “Our team has run a nightly batch job for six years that rebuilds our entire search index from scratch, taking about 4 hours. Everyone just says 'that's how the search vendor recommends it.' Can you use first principles to challenge that? Strip it down to what an index actually needs to be correct and fresh, and tell me which assumptions are real constraints and which are just habit.” **example-2** “We're a 12-person startup and we hold a 90-minute all-hands every Monday where each team presents slides. People are bored and I keep asking myself why we do it this way. Rethink it from scratch: what fundamental problem is this meeting supposed to solve, what are the basic truths about how information moves in a small company, and what would a design look like if we started from nothing?” ## Install Download the .zip and unzip it. Put `first-principles-thinking.md` here: Claude Code · `~/.claude/agents/` · Every project. For one project only, use .claude/agents/ inside it. It gives the agent the tools it asks for, and no others. Cursor · `~/.cursor/agents/` · Cursor also reads the Claude Code folder above. It has no tools field, so it gives every agent every tool, whatever this one asks for. Codex · Codex’s custom agents are TOML files, not Markdown, so this agent doesn’t install there. Keep its license file beside it, as `first-principles-thinking.LICENSE`: a name that doesn’t end in .md, since every .md file there is read as an agent. Then start a new session and ask for it by name. Folders checked 6 November 2026. Or [install from your agent](https://carteblanche.com/connect.md): ask it to install `first-principles-thinking` from Carte Blanche. ## Badge for your README Carte Blanche: Verified Paste this into the agent’s README. It links to this page, and shows how the listing stands whenever it’s loaded. ``` [![Carte Blanche](https://carteblanche.com/badge/first-principles-thinking.svg)](https://carteblanche.com/skills/first-principles-thinking) ``` ## About this test - **tested**: 7 Nov 2026 - **version tested**: v1 - **cases**: 2 - **tries per case**: 1 try per case - **run without the agent**: yes - **cases written by**: Carte Blanche - **size**: 3827 characters What this shows, and what it doesn’t How 2 models answered these 2 requests, with and without the agent. It says nothing about other models, and it doesn’t limit what your own agent can do. The answers were written without tools: the agent’s instructions were the system prompt, and it read no file and ran nothing. [How we test](https://carteblanche.com/how-it-works.md)