Skip to main content
This cheat sheet cross-references six configuration concepts across major AI tools: personalization instructions, project rules, memory, privacy opt-out, automation, and MCP. Details and sources are in the corresponding units; this table only tells you where each setting lives. Verified as of 2026-06; these tools update frequently, so confirm entries marked “needs source verification” against current official documentation before acting.
Config file names and folder layouts change. The cheat sheet gives you an alignment framework, not a list you can copy verbatim forever.

1. How to read this table

Reading across a row shows where the same concept lives in different tools. Reading down a column shows all six configuration dimensions for one tool. The six concepts follow the layer model from 02-1 Configuration Layer Model:
  1. Personalization (account-level instructions): standing instructions for how the model talks to you.
  2. Project rules: what the model should follow inside this repo.
  3. Memory: what to remember across conversations.
  4. Privacy opt-out: whether your conversations are used for training.
  5. Automation: reusable commands, skills, and event hooks.
  6. MCP: how external tools and data sources are connected.
When evaluating a new tool, work through these six dimensions: where is each one configured, and what is the default? Any cell you cannot fill in is a gap in your reading.

2. Personalization and project rules file locations

Windows paths use C:\Users\<user>\; macOS / Linux equivalent is ~.
ToolPersonalization (user level)Project levelPath scoping rulesPrivacy opt-out
Claude.ai (web)Instructions for Claude (Settings)Project instructionsEach Project is independentHelp Improve Claude
Claude Code~/.claude/CLAUDE.md, ~/.claude/settings.json./CLAUDE.md, .claude/rules/*.mdMultiple layers merged nearest-firstAccount-level Help Improve Claude
Claude CoworkNo file system (goal-oriented GUI)NoneNoneControlled by Claude account layer
ChatGPTCustom InstructionsProjects instructionsEach Project is independentImprove the model for everyone
OpenAI Codex~/.codex/AGENTS.md (global project instructions)./AGENTS.md, .codex/config.toml (loaded only when trusted)From git root down to cwd; closer files load later and overrideData Controls training toggle
Gemini (consumer)Instructions for Gemini, Personal contextGem InstructionsEach Gem is independentKeep Activity
Google Antigravity~/.gemini/GEMINI.md./GEMINI.md, ./AGENTS.md, .agents/rules/*.mdGlobal rules + workspace .agents/rules/ (plural is official default; singular compatible with older versions)Enable Telemetry (Settings → Account)
GitHub Copilot CLI~/.copilot/copilot-instructions.md.github/copilot-instructions.md, .github/instructions/*.instructions.mdPath-scoped via frontmatter applyTo globAccount / enterprise settings
GitHub Copilot Chat (VS Code)chat.* workspace settings.github/copilot-instructions.mdControlled by chat.instructionsFilesLocationsVS Code account settings
CursorUser Rules (GUI).cursor/rules/*.mdc, AGENTS.mdOnly .mdc files are read; .md is ignoredAccount settings
Cursor’s two ignore files.cursorignore blocks AI access entirely; .cursorindexingignore only excludes indexing while the file remains readable. Putting secrets in the wrong file gives you a false sense of protection.

3. Memory locations

ToolMemory file / mechanismScopeHow to disable
Claude Code~/.claude/projects/<repo>/memory/ and auto-summarized MEMORY.mdPer-repo, accumulates automaticallySettings → Memory, or instruct in CLAUDE.md not to write
Claude.aiProject conversations + non-Project conversations maintained separatelyAccount-sideSettings → Memory
ChatGPTSaved Memories (viewable and deletable per entry) + Reference chat historyAccount-side; Projects can isolateSettings → Personalization → Manage memories
GeminiPersonal context → MemoryAccount-sideSettings → Personal context → Memory
CodexNo dedicated memory; uses AGENTS.md as a standing-instructions substituteProject layerNot applicable
AntigravityMemory (account-side)Account-sideAccount settings
Copilot CLINo dedicated memory; uses instruction files as substitutePrimarily project layerNot applicable
CursorNo dedicated memory; uses .cursor/rules/*.mdc as substituteProject layerNot applicable
Same name does not mean same semanticsClaude “Memory,” ChatGPT “Saved Memories,” and the others differ in when they trigger, whether entries can be deleted individually, and whether they are used for training. Alignment as a concept is fine; assuming identical semantics will cause you to misjudge both privacy and behavioral boundaries.

4. Automation and integration locations

ToolReusable workflowsEvent hooksSub-agentsMCP config
Claude Code (primary reference).claude/skills/<n>/SKILL.mdhooks in .claude/settings.json, or standalone hooks.json.claude/agents/*.md.mcp.json (project), ~/.claude.json (user)
ChatGPTCustom GPTs (Configure tab)(no file system)(no file system)Via Custom GPT Actions
OpenAI Codexprompts/*.md (custom commands, needs source verification)[hooks] in config.toml (stable)[agents] (subagents, stable)[mcp_servers.<id>] in config.toml
Gemini (consumer)Gems(no file system)(no file system)Via Extensions
Google AntigravitySkills: IDE ~/.gemini/config/skills/, CLI ~/.gemini/antigravity-cli/skills/, workspace .agents/skills/; Workflows created via IDE panel (on-disk path not listed in official docs)hooks.json (in .agents/ or ~/.gemini/config/)Primarily AGENTS.md and .agents/~/.gemini/config/mcp_config.json (officially confirmed; CLI workspace: .agents/mcp_config.json)
GitHub Copilot CLISkills: ~/.copilot/skills/<n>/SKILL.md, .github/skills/<n>/SKILL.mdHooks: ~/.copilot/hooks/ + .github/hooks/ (additive)Agents: ~/.copilot/agents/<n>.agent.md, .github/agents/<n>.agent.md~/.copilot/mcp-config.json, .mcp.json / .github/mcp.json / .vscode/mcp.json
Cursor.cursor/rules/*.mdc(no equivalent mechanism)(no equivalent mechanism).cursor/mcp.json
Copilot hooks are additive, not overridingThe user-level ~/.copilot/hooks/ and the project-level .github/hooks/ both run together; the project layer does not replace the user layer. Before working in a shared repo, check which hooks your local setup will bring in.Antigravity is known to conflict when both Gemini CLI and Antigravity write to ~/.gemini/GEMINI.md (see 02-6 Section 3 warning).

5. Privacy opt-out quick reference

ToolToggle location (as of 2026-06)DefaultOne-off handling
Claude.ai (Free)Settings → Privacy → Help improve ClaudeOn (must manually turn off)Temporary Chat
Claude.ai (Team / Enterprise)Governed by commercial termsNo training (per terms)Terms level
Claude CodeAccount-level Help Improve ClaudeSame as aboveAccount level
ChatGPTSettings → Data Controls → Improve the model for everyoneOn (must manually turn off)Temporary Chat
GeminiSettings → Keep ActivityOn (users 18+)Conversations after disabling
Workspace (Gemini)Administrator-controlledDetermined by org policyNot applicable
CopilotAccount / enterprise settingsEnterprise accounts default to no trainingPersonal accounts follow account settings
CursorAccount Privacy settingsDepends on account typePrivacy Mode (version-dependent)
Defaults are a trapConsumer plans (ChatGPT, Gemini Free) typically default to using conversations for model improvement; you have to opt out yourself. Do not assume enterprise defaults apply to personal accounts, or vice versa.

6. Cross-tool rules baseline: AGENTS.md

AGENTS.md is an open standard for converging each tool’s proprietary rules file into one (governed by the Agentic AI Foundation under the Linux Foundation). Adoption: Codex, Cursor, Copilot, Gemini CLI, Jules, Devin, Amp, VS Code, and over twenty other tools support it natively (as of 2026-06, adopted by more than sixty thousand open-source projects).
ToolNative supportHow to share
Claude CodeNo (does not read AGENTS.md)Import via @AGENTS.md syntax in CLAUDE.md
OpenAI CodexYesReads root and all parent layers by default
Google AntigravityYesUsed alongside GEMINI.md
GitHub Copilot CLIYesRoot is primary; can serve as substitute for CLAUDE.md / GEMINI.md
CursorYesUsed alongside .cursor/rules/*.mdc
AGENTS.md is a baseline, not a replacementDecide which file is the baseline (usually AGENTS.md) and have the others point back to it. Duplicated content drifts; eventually every copy is half-wrong.

7. Hands-on: 30-second alignment exercise

Scenario: migrating a Claude setup to Codex
1

Fill in the six-concept table for your current Claude setup as the source of truth

  • Personalization: ~/.claude/CLAUDE.md
  • Project rules: ./CLAUDE.md, .claude/rules/
  • Memory: auto-memory + MEMORY.md
  • Privacy opt-out: Help improve Claude toggle
  • Automation: skills / hooks / commands
  • MCP: .mcp.json (project), ~/.claude.json (user)
2

Map each cell to Codex

  • Personalization: ~/.codex/AGENTS.md
  • Project rules: ./AGENTS.md, .codex/config.toml
  • Privacy opt-out: Data Controls training toggle
  • Automation: [hooks] in config.toml, prompts/*.md, [mcp_servers.<id>]
  • MCP: same as automation
  • Memory: none dedicated; use AGENTS.md as substitute
3

Extract overlapping project rules into AGENTS.md so both tools share one source

Once you do this, you will find only two real differences: Codex centralizes configuration in config.toml, while Claude uses settings.json.

8. Corresponding units