End-to-end baseline 2026-04-28
Snapshot of the full ingest → dbt → frontend pipeline taken right after PLAN-003 merged. This is the known-good starting point for further development. Re-run the same sweep before any major refactor to catch regressions.
AGENT runbook — onboard a new data source
This is the autonomous-agent runbook for adding one upstream data source to Atlas's ingest pipeline. It is written to be read by a Cursor Background Agent (or similar cloud agent) running in a sandbox VM with a fresh repo clone. Humans should read contributors/adding-a-source.md instead — that is the canonical 11-step workflow, written for humans. This file is the agent-shaped projection of the same workflow, with explicit invariants, gates, and escalation paths.
Working Inside the DevContainer
All projects use the DevContainer Toolbox (DCT) for development. The AI must run all commands inside the devcontainer, never on the host machine.
Git Safety Rules and Operations
Git operations require user confirmation. The AI must never run destructive git commands without explicit approval.
Implementation Plans
How we plan, track, and implement features and fixes.
AI Developer Guide
Instructions for AI coding assistants working on this project.
Talk — AI-to-AI Testing Protocol
Talk is a file-based communication protocol that enables two separate AI coding sessions to collaborate on testing. One session develops and builds, the other tests as a fresh user. They communicate by appending messages to a shared talk.md file.
Plan to Implementation Workflow
How ideas become implemented features.
Running Two Claude Sessions in the Same Repo
Running multiple Claude Code (or Claude VS Code) sessions against the same
plans
2 items
Project: Atlas
Atlas is an organisation-neutral information platform that aggregates public data about every large Norwegian NGO. The product surface is a Next.js App Router web app at atlas.helpers.no (TypeScript, React Server Components, Digdir Designsystemet for UI, MapLibre for maps). The data behind it is produced by a separate pipeline (atlas-data/) that ingests Norwegian public-sector sources (SSB, FHI, Brreg, Kartverket, Bufdir, IMDi, NAV, Lottstift, Innsamlingskontrollen, …), transforms them through dbt, and serves them as marts.* tables in PostgreSQL.