Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/nasqret/lean-interact/claude-mdgit clone --depth 1 https://github.com/nasqret/lean-interactWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/nasqret/lean-interact/claude-md)<a href="https://agentmods.dev/instructions/nasqret/lean-interact/claude-md"><img src="https://agentmods.dev/badge/instructions/nasqret/lean-interact/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.04212 | $0.04212 |
| Opus 5 | $0.02106 | $0.02106 |
| Sonnet 5 | $0.00842 | $0.00842 |
| Haiku 4.5 | $0.00421 | $0.00421 |
Grade A, and why
lean-interact CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 298 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — operating instructions for this repository
This file tells the assistant how to work in lean-interact. It is not background reading;
it is the procedure. Follow it literally.
The short version: the user states mathematics, you produce compiled Lean, fast, and you never assert a Mathlib name you have not verified. Everything below serves those three things.
0. Start of every session
Before responding to the first mathematical request, read, in this order:
MEMORY.md— durable facts and preferences. Non-negotiable context. It tells you who the user is, what the environment guarantees, and which rules have been established. If its template sections are still unfilled, assume the default this project was built for: a working mathematician who knows the mathematics and is learning Lean and Mathlib. Then ask the user to fill those sections in, once, and record the answers.vault/Formalization Style.md— how Lean is written here: naming, statement shape, tactic style,ℕvsℤvsZMod n. Entries marked PROVISIONAL are the project's guess, not the user's stated wish; follow them, but never defend them. If the user objects, the user is right and the note gets updated.
Then skim the top entry of JOURNAL.md for where the last session stopped.
Do not skip this because the request looks simple. The whole point of those files is that they contain decisions you would otherwise re-litigate or silently contradict.
If a live session is not already running, start it: tools/session.sh (see --help).
Five skills package the recurring procedures, and you should use them rather than
improvising: install (bring a fresh clone to a working state — the precondition for all
the others), formalize (prose in, theorem out), formalize-from-magma (computation
in, theorem out), mathlib-lookup (find and verify a name), lean-session (bring up
or repair the live views).
1. The formalization loop
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 298 lines · 4,212 tokens per session scan A aab1c2cf7576
lean-interact CLAUDE.md is an instructions file published in the GitHub repository nasqret/lean-interact (10 stars, last pushed 26d ago), licensed MIT. It adds 4,212 tokens to every session, about $0.0211 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
formal CLAUDE.md
Instructions for yamafaktory/formal, covering formal, rust, the files that judge changes, hints and lean.
leanforge-mcp AGENTS.md
AGENTS.md instructions for sandraschi/leanforge-mcp, covering agents.md -- leanforge-mcp, stack, repo layout, critical rules and lean subprocess.
leanforge-mcp CLAUDE.md
Claude Code instructions for sandraschi/leanforge-mcp, covering claude.md -- leanforge-mcp, what this repo does, key concepts, when working on agent.py and when working on leanclient.py.
LeanProbe AGENTS.md
Instructions for epfl-lara/LeanProbe, covering agents.md, working on this repo and the bundled skill.
VCVio AGENTS.md
AGENTS.md instructions for Verified-zkEVM/VCVio, covering vcvio — ai agent guide, fast start, attribution, headers, and docstrings, module scopes and what this project is.
jacobian AGENTS.md
AGENTS.md instructions for morluto/jacobian, covering jacobian agent guide, what we are building, non-negotiable boundaries, implement mathematics directly and mathematical boundedness is a proof obligation.