Command Equivalents · AI coding agents

How to do the same task in each tool. All commands below are verified from the official documentation.

Task	Claude Code	Codex	Kimi Code
Start session	`claude` (or `claude "query"`)	`codex` (or `codex "PROMPT"`)	`kimi`
Resume session	`claude -c` (most recent in cwd); `claude -r <id-or-name>`; `/resume`	`codex resume [SESSION_ID]` or `codex resume --last`; `/resume`	`kimi -c` (most recent in cwd); `kimi -S [id]`; `/sessions`
Clear/new context	`/clear [name]`	`/new [name]` or `/clear` (clears terminal and starts a fresh chat)	`/new` (alias `/clear`)
Change model	`/model [model]`; `claude --model <model>`	`/model`; `codex -m <model>`	`/model`; `kimi -m <model>`
Compact context	`/compact [instructions]`	`/compact`	`/compact [instruction]`
Plan before coding	`/plan [description]`; Shift+Tab; `claude --permission-mode plan`	`/plan [inline prompt]`	`/plan [on\|off]`; Shift-Tab; `kimi --plan`
Review changes / code review	`/code-review [--fix] [target]`; `/review [PR]`; `/security-review`	`/review` (working tree); `codex review --uncommitted \| --base <branch> \| --commit <SHA>`	No direct equivalent verified.
Change permissions	`/permissions`; Shift+Tab cycles modes	`/permissions` (approval presets)	`/permission`; `/yolo`; `/auto`
Add MCP server	`claude mcp add --transport http <name> <url>`; `claude mcp add <name> -- <command>`	`codex mcp add <name> -- <command>`; `codex mcp add <name> --url <url>`	`/mcp-config` (interactive); edit `~/.kimi-code/mcp.json` or `.kimi-code/mcp.json`
Non-interactive run	`claude -p "query"`	`codex exec "PROMPT"`	`kimi -p "prompt"`
Generate project instructions file	`/init` → CLAUDE.md	`/init` → AGENTS.md	`/init` → AGENTS.md
Undo/rewind	`/rewind` (code + conversation checkpoint rollback; also Esc Esc)	Esc Esc (empty composer) edits previous message and forks	`/undo [count]` (prompts, todo list, plan state; not code); Esc Esc opens undo selector

Behavior differences that matter

claude -c vs codex resume --last vs kimi -c — All three continue the most recent session scoped to the current working directory. Differences: codex resume without flags opens an interactive session picker (use --last for most recent, --all to search sessions from any directory), and if the cwd differs from the session’s saved directory Codex asks which directory to use. kimi -c and claude -c go straight to the most recent session in the cwd. Kimi’s resume restores the bound agent and the saved permission/plan mode (mode flags may override — the docs conflict on whether --yolo/--plan combine with --continue; treat as unresolved). Claude additionally supports --fork-session (resume into a new session ID) and --from-pr (resume the session linked to a GitHub PR).
/compact — Available in all three, but the trigger and knobs differ. Claude auto-compacts at ~95% of the context window (CLAUDE_AUTOCOMPACT_PCT_OVERRIDE to adjust) and fires PreCompact hooks with matcher manual/auto. Codex auto-compacts at model_auto_compact_token_limit and the summary prompt is customizable via compact_prompt. Kimi auto-compacts based on [loop_control] reserved_context_size; after a compaction, prompts from before it cannot be undone. Claude and Kimi accept an optional instruction to guide what the summary preserves; the Codex docs do not describe an inline instruction for /compact.
Clear vs new — In Claude, /clear clears conversation history (with optional name). In Codex, /new starts a new chat in the same CLI session while /clear both clears the terminal and starts a fresh chat (Ctrl+L only clears the view). In Kimi, /new and /clear are aliases that start a fresh session discarding the current context.
Review — Claude’s three review commands have different semantics: /code-review is a multi-agent diff review with depth levels (up to ultra, which runs in the cloud), /review is a fast single-pass read-only GitHub PR review, and /security-review checks the diff for vulnerabilities. Codex’s /review covers the working tree in the TUI, while codex review runs non-interactively against uncommitted changes, a base branch, a commit, or a custom prompt; it does not modify the working tree. Kimi Code has no documented code review command.
Undo/rewind is not the same feature in each tool — Claude’s /rewind rolls both code and conversation back to a checkpoint (and can resume from before a /clear). Kimi’s /undo rolls back prompts, the todo list, and plan-mode state, but explicitly not code changes. Codex has no undo command; Esc Esc with an empty composer lets you edit a previous message, which forks the chat.
Non-interactive defaults — claude -p supports --output-format text|json|stream-json, --max-turns, --max-budget-usd, and --permission-prompt-tool; built-in/user slash commands are not available in this mode. codex exec defaults to a read-only sandbox, streams progress to stderr and prints only the final message to stdout, and requires a Git repo unless --skip-git-repo-check is passed. kimi -p runs with the auto (Never Ask) permission policy by default — no human approval — though static deny rules still apply; output formats are text and stream-json.
MCP server registration — Claude and Codex both have CLI subcommands (claude mcp add, codex mcp add) plus in-session panels (/mcp). Kimi has no documented CLI add command; servers are configured interactively via /mcp-config or by editing mcp.json files directly.

Last verified: 2026-09-07

Official sources:

AI coding tools evolve quickly. If you find outdated information, please open an issue or pull request.