“Cursor | .cursor/rules/*.mdc, AGENTS.md | Supports always-on, intelligently applied, manually invoked, and file-scoped rules.”
“Cursor has a particularly convenient rules interface.”
“Excellent for editor-centric development”
“Cursor also supports AGENTS.md, which makes it easier to share basic instructions with other coding agents.”
“The main concern is that intelligently applied rules are not necessarily included in every relevant request.”
“GitHub Copilot | .github/copilot-instructions.md, AGENTS.md | Supports repository-wide and path-specific instructions, including nested AGENTS.md.”
“Copilot | Broad instruction integration | Good, but verify the active harness”
“This makes Copilot flexible, but it is worth checking which instruction sources are actually active in your particular environment.”
“An important detail: the actual agent harness determines which instructions are loaded.”
“Windsurf | .windsurfrules, AGENTS.md support varies by integration | Uses persistent memories and project rules to guide Cascade.”
“Windsurf has project rules and Memories, but its interesting differentiator is its hook system.”
“Interesting for automated workflow enforcement”
“One caveat: a hook triggering does not prove that the resulting code complies with your conventions.”
“OpenAI Codex | AGENTS.md | Automatically discovers instructions along the project directory hierarchy; supports more-specific overrides.”
“Codex uses AGENTS.md as its primary project instruction format.”
“Excellent for shared repository conventions”
“This is particularly useful in monorepos.”
“Claude Code | CLAUDE.md, .claude/rules/ | Loads project guidance, supports nested instructions and path-specific rules.”
“Claude Code offers several mechanisms with different purposes:”
“Excellent for complex, detailed workflows”
“One advantage is that instructions can be organized without putting everything into one enormous context file.”