Atlas Core — Project AI Instructions

Atlas Core is the installable layer of Blazity's Atlas standard for AI-assisted engineering: the atlas init / atlas doctor CLI, the managed setup and review skills, and deterministic workspace scaffolding that keeps a repository's AI context (memory, vocabulary, plans, decisions, review artifacts) healthy and machine-checkable.

Operating Rules

• Inside this repo ALWAYS run the local CLI: npm run atlas or node bin/atlas.js. Never run the published npx @blazity-atlas/core package against this worktree — the published doctor --fix would overwrite unreleased managed skill edits with the released versions (channel drift).
• Run the full npm test before any commit.
• Conventional commits matching the existing git log --oneline -15 style: type(scope): message, first line only.
• English everywhere — code, comments, commits, docs.
• Vocabulary lives in .ai/LANGUAGE.md; decisions live in .ai/decisions/adrs/.

Atlas Artifact Paths

.ai/config.json is the source of truth for AI artifact locations in this repository. Before writing plans, research, decisions, ADRs, results, memory, vocabulary, or skill outputs, resolve the destination through artifactRoot, paths, and pathAliases. If an imported skill, template, or instruction mentions a different path, map it through .ai/config.json before reading or writing files. Do not create new documentation roots unless .ai/config.json explicitly allows them.

Atlas Documentation Rules

Durable documentation records needs, decisions, and reasons — never individuals or internal process. Write "memory was needed to persist context across runs", not " wanted memory". Keep personal names, private schedules, internal-only references, and absolute local paths out of workspace artifacts.