從 Cursor 搬家到 Claude Code / Codex：遷移盤點 + 起步設定 prompt

You are helping me migrate this repository from Cursor to {{TARGET_TOOL}} (either Claude Code or OpenAI Codex CLI). Do NOT make any code changes — this is a configuration migration only. Work in two phases and stop after Phase 1 for my approval.

## Phase 1 — Inventory (read-only)
Scan the repo for existing Cursor configuration and report what you find. Look specifically for:
- `.cursorrules` (single-file rules)
- `.cursor/rules/*` (rule files / project rules)
- `.cursor/mcp.json` (MCP server definitions)
- Any `.cursorignore` file
- Project conventions implied by the code that AREN'T written down anywhere (e.g. test command, lint/typecheck command, package manager, framework, directory structure, naming conventions)

Produce an inventory table: `Source file | What it configures | Where it maps to in {{TARGET_TOOL}} | Notes`.

Mapping rules to apply:
- `.cursorrules` and `.cursor/rules/*` → persistent project memory: `CLAUDE.md` for Claude Code, `AGENTS.md` for Codex (at repo root).
- Reusable command-shaped rules ("always do X when I ask for Y") → custom slash commands under `.claude/commands/<name>.md` (Claude Code) — Codex has no equivalent slash-command file, so fold these into AGENTS.md sections instead.
- `.cursor/mcp.json` servers → MCP config: for Claude Code suggest the `claude mcp add` commands; for Codex put them under `[mcp_servers]` in `~/.codex/config.toml`.
- `.cursorignore` → note that the agent already respects `.gitignore`; flag anything extra that needs documenting.

List anything in the Cursor setup that does NOT cleanly map (Cursor-editor-specific features) and say so honestly instead of inventing an equivalent.

## Phase 2 — Generate the new config (after I approve)
For {{TARGET_TOOL}}, write the actual files:
1. A complete `{{MEMORY_FILE}}` (CLAUDE.md or AGENTS.md) with these sections: **Project overview** (1 short paragraph), **Tech stack**, **Commands** (exact build / test / lint / typecheck commands you discovered or that I provide: {{KEY_COMMANDS}}), **Code style & conventions**, **Architecture / directory map**, **Do / Don't** (gotchas). Convert any Cursor-specific phrasing into tool-neutral instructions. Keep it tight — bullet points over prose.
2. If Claude Code: draft `.claude/commands/` files for each reusable workflow you found, with frontmatter `description`.
3. The MCP migration commands or config block.

Then give me a short "day-one workflow note" explaining the mindset shift: in Cursor I approve each edit inline; in {{TARGET_TOOL}} I should let a batch of edits run and review the resulting `git diff` at the end, and use plan mode / read-only exploration first on anything non-trivial.

Do not fabricate features. If you are unsure whether {{TARGET_TOOL}} supports something, say "unsure — verify in docs" rather than guessing.

把方括號 [ ] 內的變數換成你的內容，丟進 Claude Code / Codex。