# Project instructions

## Start of each session
Read PROJECT.md and HANDOFF.md from this repository before continuing. Treat these files as the shared project knowledge across computers and conversations. Verify the latest repository version before editing.

## Working approach
Use AtomHMI's graphical web interface to create and edit pages when requested. Do not assume that this repository contains the AtomHMI application source. Ask for the target site, page, and desired change when they are missing.

Record confirmed facts separately from assumptions. Do not invent page structures, addresses, credentials, or completed work. Never commit passwords, tokens, cookies, or other credentials.

## Continuity
At the end of a substantial work session, update HANDOFF.md with completed work, validation results, unfinished items, and concrete next steps. Update PROJECT.md when durable project facts or decisions change. Commit these updates to this repository and report the commit. Do not claim knowledge was saved until the remote write succeeds.

Repository notes provide continuity, not automatic synchronization of chat history, browser sessions, running tools, or local files. A new chat must explicitly read them.

## Atom HMI site documentation instructions

# Atom HMI project knowledge

The user requests ongoing documentation upkeep as part of future Atom HMI work.

Before site work, read the private cloud knowledge base and relevant site guides:
https://chatgpt.com/space/page_59b26c8716dc8191a88caea787825299

Cloud Page IDs and local filenames are recorded in cloud-knowledge-base.json. Read current cloud content before updating it; reconcile remote edits rather than overwriting them with an older local copy.

During each authorized site task, update the relevant cloud guide with verified new knowledge and completed changes without waiting for a separate request. Create and index a guide for any newly explored site. Update ATOMHMI-PLAYBOOK.md and its cloud Page for techniques that apply across sites. Keep local Markdown backups aligned when available.

Document site/environment/device IDs, equipment and register mappings, units/conversions, writable settings versus outputs, page patterns, alarms/latching, trends, tests, restored values, and unresolved questions. Distinguish observations, user-confirmed facts, and assumptions. Correct stale statements with dated evidence while preserving useful history.

Never store credentials. Respect current site authorization; simulation permission for one site does not apply automatically elsewhere. Report any failed documentation save. This is upkeep during active project tasks, not authorization to create a scheduled automation.