rMax.AI is an agent-first research lab focused on building, evaluating, and publishing reproducible research about AI-native systems and agent architectures. This repository contains the public site, research notes, agent prompts and minimal operational artifacts used to run, audit, and iterate on agent-driven workflows.

• Agent-first engineering: treat agents as modular specialists that can be composed, tested, and audited.
• Authority-first design: give agents clear, scoped authority and explicit success criteria to reduce risky open-ended behavior.
• Failure-oriented orchestration: build workflows that detect, contain, and recover from partial failures.
• Earned autonomy: grant more responsibility to agents only after human-reviewed performance and safety checks.
• Transparency & reproducibility: store prompts, drafts, and publish artifacts in the repo so research is auditable and repeatable.

• .agent/: agent and mode definitions describing responsibilities and orchestration.
• .agent/prompts/: reusable prompt templates for writing, reviewing, and publishing notes.
• inbox/: captured ideas and raw inputs that feed agent workflows.
• processed/: agent-produced drafts and intermediate artifacts.
• notes/: published notes and essays (public output).
• research/: longer-form projects, experiments and notebooks.
• scripts/: helper scripts (e.g., scripts/generate-sitemap.py).
• docs/: supporting documentation and operational playbooks.
• images/: site and asset images.
• index.md / index.html: static site entry points.

1. Idea captured in inbox/.
2. Orchestration agent decomposes the idea into tasks and assigns subagents.
3. Subagents synthesize drafts, run experiments, or generate artifacts and drop them into processed/.
4. Human reviewers use review prompts (see .agent/prompts/) to audit and iterate.
5. After approval, artifacts are published to notes/ and the site is updated.

1. Add a mode / metadata file in .agent/mode/.agent.md describing inputs, outputs, responsibilities and failure modes.
2. Add or update prompt templates in .agent/prompts/ (use existing prompts as examples).
3. Prefer small, testable behaviors with explicit success criteria and recovery paths.
4. Test changes by running the agent orchestration in a controlled environment or via manual prompt simulations.

• This is a static site. Quick local preview:
  • python -m http.server 8000
  • Open http://localhost:8000
• Use scripts/ for maintenance tasks such as sitemap generation.

• Open issues or PRs on GitHub: https://github.com/rmax-ai/rmax-ai.github.io
• Add new ideas to inbox/ to feed agent workflows.
• Propose changes to modes and prompts with examples and expected outputs.

• All agent actions producing public artifacts go through human review before publication.
• Failure-mode reviews and review prompts are first-class artifacts in .agent/prompts/.
• The project prioritizes reproducibility, auditability and clear, testable agent behavior.

• Website: https://rmax.ai
• Repo: https://github.com/rmax-ai/rmax-ai.github.io
• Issues: https://github.com/rmax-ai/rmax-ai.github.io/issues

• No license specified in this repository. Add a LICENSE file or contact the project owner if you need reuse terms.