create-slash-commands

<quick_start>

1. Determine command purpose and whether it needs user arguments
2. Create .claude/commands/command-name.md (project) or ~/.claude/commands/command-name.md (personal)
3. Add YAML frontmatter (description required; argument-hint, allowed-tools, model optional)
4. Write XML-structured body with objective, process, success_criteria
5. Test with /command-name [args] </quick_start>

<essential_principles> Commands are instructions FOR Claude, not messages TO users. When a user invokes /command-name, the file content becomes Claude's prompt. Write directives telling Claude what to do.

Correct: "Review this code for security vulnerabilities and report findings with line numbers." Wrong: "This command will review your code for security issues." </essential_principles>

<step_2_frontmatter> Build YAML frontmatter:

---
description: Clear, actionable description (under 60 chars)
argument-hint: [input-description]          # Only if command needs arguments
allowed-tools: Read, Grep, Bash(git:*)      # Only if restricting tools
model: sonnet                               # Only if overriding model
disable-model-invocation: true              # Only if manual-only
---

Fields:

• description (required) -- shown in /help listing
• argument-hint -- documents expected arguments for autocomplete
• allowed-tools -- restricts available tools (string, array, or Bash patterns)
• model -- forces specific model (sonnet, opus, haiku)
• disable-model-invocation -- prevents programmatic invocation </step_2_frontmatter>

<step_3_body> Write the command body using XML tags:

Always include:

• <objective> -- what and why
• <process> -- numbered steps
• <success_criteria> -- definition of done

Include when relevant:

• <context> -- dynamic state via bash execution or file references
• <verification> -- checks to perform after creating artifacts
• <testing> -- test commands to run
• <output> -- files created or modified </step_3_body>

<step_4_arguments> Handle arguments based on command type:

No arguments (self-contained): Omit argument-hint, don't reference $ARGUMENTS. Pattern: Command operates on implicit context (current conversation, known files, project state).

$ARGUMENTS (single string): All user input as one string.

Fix issue #$ARGUMENTS following project conventions.

Usage: /fix-issue 123 high-priority

Positional ($1, $2, $3): Structured input.

Review PR #$1 with priority $2 and assign to $3.

Usage: /review-pr 456 high alice

See references/argument-patterns.md for advanced patterns. </step_4_arguments>

<step_5_dynamic_context> Load runtime context into commands:

Bash execution (output included in prompt): Note: In actual commands, place ! directly before the backtick with no space.

Current status: ! `git status`
Recent commits: ! `git log --oneline -5`

File references (content loaded): Note: In actual commands, place @ directly before the path with no space.

Review @ src/api/users.ts for security issues

Plugin resources:

Template: @ ${CLAUDE_PLUGIN_ROOT}/templates/report.md
Script output: ! `bash ${CLAUDE_PLUGIN_ROOT}/scripts/analyze.sh`

</step_5_dynamic_context>