3 ms·
Show HN: Llms.md – a README.md for LLM-powered coding agents
Hi HN,
We just published LLMS.md, an open specification for helping Large Language Models (LLMs) understand code repositories.
The idea is simple:
• README.md helps humans
• robots.txt helps crawlers
• LLMS.md helps AI coding agents (Copilot, Cursor, Claude Code, Gemini CLI, etc.)
A root-level LLMS.md gives agents:
• Key directories & entry points
• Which files to ignore (dist, deps)
• Run/test instructions
• Reasoning hints from maintainers
Spec: github.com/llmspec/llms-spec
RFC: llmspec/llms-spec/rfc/0001-llms-md-spec.md
We’d love feedback, contributors, and early adopters. If you maintain an OSS repo, try adding a LLMS.md and let us know what worked!
- attogram 1y agoUh.. you mean AGENTS.md ?
- huan42 1y agoYep, AGENTS.md exists Our goal with LLMS.md is narrower: think of it as the repo twin of llms.txt. AGENTS.md says “here’s how any agent should behave here.” LLMS.md says “here are the files/dirs/tests that actually matter — in a format machines can trust.” Services like gitingest show the demand: everyone wants a flattened, LLM-ready repo string. LLMS.md just gives those tools a canonical upstream spec instead of guessing.
- huan42 1y agoAnticipating a common question: “Uh… you mean AGENTS.md?” Short answer: • AGENTS.md = broad, freeform guidance for any agent (already supported by Copilot). • LLMS.md = structured, schema-like spec tailored for LLMs (dirs, entry points, ignore, run/test). They’re complementary, not competing. Think of it as: • AGENTS.md → notes for agents (“how to behave here”) • LLMS.md → metadata for models (“what matters in this repo”) Why LLMS.md? We’re aligning it with llms.txt on the web. Websites have llms.txt to guide LLMs to important pages. Repos now get LLMS.md to guide LLMs to important files. Together, they form a predictable convention across ecosystems. Where tools fit in: Services like gitingest already flatten repos into an LLM-readable string. LLMS.md gives them an upstream spec so they don’t have to guess what to include or ignore. Big picture: • llms.txt → websites • LLMS.md → code repos • gitingest & friends → renderers that consume the spec • Agents/IDEs → benefit from curated, less noisy context That’s the ecosystem we’d like to help build. Feedback welcome!
- attogram 1y agoI'm currently doing both tasks in AGENTS.md - both how to behave, and specific details on what matters in the repo. What are the advantages of splitting this into 2+ files? Isn't it just more overhead?