ORE Studio as an LLM-first system

Table of Contents

Summary

ORE Studio is an LLM-first system: large language models are the primary means of development, working under human direction. All documentation is written to be equally useful to human readers and AI agents. Claude Code is the primary tool, but any LLM that can follow the project's LLM instructions is welcome to contribute.

Detail

What LLM-first means

ORE Studio does not treat LLMs as a convenience layer bolted on top of a human workflow. The project is designed so that an LLM session starting from scratch can orient itself, pick up work, implement it, and raise a PR without human hand-holding at each step. The LLM instructions are the mandatory entry point; the compass tool provides live orientation; recipes, runbooks, and skills encode every repeatable operation.

Human direction, LLM execution

Humans set goals, review work, and make architectural decisions. LLMs execute: they write code, author documentation, open PRs, respond to review comments, and close stories. The agile process (sprint, story, task, PR) is the handoff boundary — a human approves a story's goal and acceptance criteria; an LLM implements and delivers it.

Documentation for two audiences

Every document in ORE Studio is written to serve both readers. For humans: clear prose, navigable headings, worked examples. For LLMs: consistent #+type: and #+description: frontmatter, org-roam :ID: properties on every node, [[id:UUID][Title]] cross-references that can be resolved by the doc graph, and executable #+begin_src sh blocks in recipes. The structured format lets an LLM load a page and immediately know what it is, where it fits, and how to act on it.

Primary tool: Claude Code

Claude Code (Anthropic's CLI for Claude) is the tool used for day-to-day development. Configuration, permissions, and session behaviour are described in Claude Code and its associated settings.

Contributing from other LLMs

ORE Studio accepts contributions from any LLM, not just Claude. The only requirement is that the LLM reads and follows LLM instructions before acting. Commit trailers identify the contributing model (Co-Authored-By: with the model name and noreply@anthropic.com / noreply@google.com etc.); see commit conventions for the exact format.

See also

Emacs 29.1 (Org mode 9.6.6)