AI Agent Entry Points As Operational Contracts

How AGENTS.md turns repository expectations into a durable operating contract for AI-assisted work.

An AI agent entry point is more than a prompt file. In these repositories, `AGENTS.md` is the operating contract for how automated help should behave inside the project.

The file names the production loop, local checks, repository guardrails, uncertainty rules, and release boundaries. Adapter files for specific tools point back to it so the repository has one canonical instruction source.

The entry point also now requires organization-level reuse. Before inventing a reusable pattern, practice, term, schema convention, prompt convention, CI convention, or release rule, agents must consult the organization patterns and glossary repositories.

That matters because portfolio production should not fork its own vocabulary by accident. A useful practice discovered in one game can become a reusable pattern. A term introduced in one repo can become a glossary entry. Article and devlog candidates can be proposed when the work is worth explaining publicly.

The result is a calmer model for AI-assisted work: read the contract, preserve uncertainty, keep records current, run checks, and propose reusable knowledge when the work creates it.