llm-cursor-source

name	llm-cursor-source
description	Repository source of truth for AI editor guidance is under .cursor, .cursorrules, and .cursorignore. Use when adding, editing, or deleting skills, rules, .cursorrules, or .cursorignore.

LLM / Cursor source of truth

What to commit

Only these paths define shared guidance for this repo:

• .cursor/skills/**
• .cursor/rules/**
• .cursor/prompts/**
• .cursor/hooks/** and .cursor/hooks.json (when present)
• .cursorrules (repo root)
• .cursorignore (repo root) — path-level ignores for Cursor

Do not commit ad-hoc skill or rule trees under .github/ or other paths outside .cursor/.

Never add or recreate .llm/exports/ — multi-editor mirror trees are prohibited. Cursor-only source under .cursor/ is the policy.

Operator directory docs (CURSOR-*.md, LLM-*.md) describe folders for humans; they are not a second source of agent rules.

When .cursor changes

• Edit skills, rules, prompts, hooks, .cursorrules, and (when needed) .cursorignore, then commit them like any other source.
• Do not add duplicate guidance elsewhere; keep one source of truth under .cursor/.
• Use an llm/<kebab-name> branch when the PR changes only LLM-related paths (.cursor/**, .cursorrules, .cursorignore, docs/development/llm/**, .llm/**, relevant AGENTS.md sections). See DOCS-DEVELOPMENT-LLM.md § Branch naming.

Skill file hygiene

• Skill frontmatter name should match the skill folder name.
• Keep exactly one blank line between closing frontmatter (---) and body content.

Contributor policy

Overview, plans, and optional history notes:

• docs/development/llm/DOCS-DEVELOPMENT-LLM.md