How commands work

Free sample

Format, porting, and what makes a command worth creating.

commands/README.md · 1.7 KB

commands/README.md

# Slash commands

A command is a prompt you invoke by name. Where a skill fires when the work
matches, a command fires when *you* say so.

## Format (Claude Code)

Any markdown file in `.claude/commands/` becomes `/filename`:

```markdown
---
description: One line, shown in the command list
argument-hint: [what to pass]
---

The prompt body. $ARGUMENTS is replaced with whatever followed the command.
```

Project commands live in `.claude/commands/` and are committed. Personal
ones live in `~/.claude/commands/` and are not.

## Porting

| Tool | Equivalent |
|---|---|
| Claude Code | `.claude/commands/<name>.md` → `/name` |
| Codex | `~/.codex/prompts/<name>.md` → `/name` |
| Cursor | No direct equivalent — keep as `docs/prompts/<name>.md` and paste |
| Cline / Roo | Workflow files, or paste |

Even without tool support these are worth keeping as files: a prompt you
have refined nine times should not live in your shell history.

## What is here

    bootstrap.md   Draft a memory file for an unfamiliar repo
    repro.md       Reproduce a bug before diagnosing it
    spec.md        Turn a vague request into a written spec
    review.md      Run a review pass on the current diff
    handoff.md     Summarise a session so a fresh one can continue
    ship.md        Pre-merge checklist

## Writing your own

The commands worth creating are the ones where you type the same three
paragraphs repeatedly. Signs you need one:

- You have a note file you copy from
- You always forget one step of a procedure
- The prompt is long enough that you shorten it and get worse results

Keep them **imperative and specific**. A command that says "help me review
this" is not worth the file.

That was one of 28 files

The rest of The Agent Operating Kit is written the same way. $39 once, yours permanently, 14-day refund.

# Slash commands A command is a prompt you invoke by name. Where a skill fires when the work matches, a command fires when *you* say so. ## Format (Claude Code) Any markdown file in `.claude/commands/` becomes `/filename`: ```markdown --- description: One line, shown in the command list argument-hint: [what to pass] --- The prompt body. $ARGUMENTS is replaced with whatever followed the command. ``` Project commands live in `.claude/commands/` and are committed. Personal ones live in `~/.claude/commands/` and are not. ## Porting | Tool | Equivalent | |---|---| | Claude Code | `.claude/commands/<name>.md` → `/name` | | Codex | `~/.codex/prompts/<name>.md` → `/name` | | Cursor | No direct equivalent — keep as `docs/prompts/<name>.md` and paste | | Cline / Roo | Workflow files, or paste | Even without tool support these are worth keeping as files: a prompt you have refined nine times should not live in your shell history. ## What is here bootstrap.md Draft a memory file for an unfamiliar repo repro.md Reproduce a bug before diagnosing it spec.md Turn a vague request into a written spec review.md Run a review pass on the current diff handoff.md Summarise a session so a fresh one can continue ship.md Pre-merge checklist ## Writing your own The commands worth creating are the ones where you type the same three paragraphs repeatedly. Signs you need one: - You have a note file you copy from - You always forget one step of a procedure - The prompt is long enough that you shorten it and get worse results Keep them **imperative and specific**. A command that says "help me review this" is not worth the file.