AGENTS.md
Conventions for AI coding agents working in the JOGL Network repository.
AGENTS.md is a file at the
repository root that tells AI coding agents how to work safely in this codebase. It follows
the emerging agents.md convention — a single, predictable place for
agent-facing instructions — and mirrors the repository's CLAUDE.md.
This page is about the repo-development AGENTS.md (how an agent should modify the
code). It is different from the MCP server and
REST API, which are how an agent uses the product at
runtime.
What it covers
AGENTS.md documents the things an agent needs to avoid breaking production, including:
- Hosting & deployment — the three environments (prod/staging/dev), each a Vercel project paired with a Supabase project, and how deploys are triggered.
- Tech stack — React + Vite + TypeScript frontend, Supabase (Postgres, Edge Functions, RLS, pg_cron) backend, and the AI/ML providers.
- Architecture rules — PostgREST footguns (
max_rowstruncation,.in()URL limits,.maybeSingle()hazards), Edge Function and migration conventions, and state-management boundaries (Zustand vs. React Query). - High-risk areas — auth, RLS, streaming/SSE, the search pipeline, and the enrichment pipeline's phase transitions.
- Scope discipline — make the smallest reasonable change, don't refactor unrelated code, and don't touch schema/RLS/auth without explicit instruction.
Using it
If you're pointing a coding agent (Claude Code, Copilot Coding Agent, or another) at this
repo, it should read AGENTS.md first. The file is the authoritative, living description of
how to make changes here — treat its rules as overriding defaults.