List Commands

List all custom slash commands from the current repository using efficient single-pass extraction.

Step 1: Detect Repository Type

Determine which command directories to search:

# Check for marketplace repo
if [ -f "marketplace.json" ] || [ -f ".claude-plugin/marketplace.json" ]; then
    echo "REPO_TYPE=marketplace"
fi

# Check for plugin repo
if [ -f ".claude-plugin/plugin.json" ] && [ -d "./commands" ]; then
    echo "REPO_TYPE=plugin"
fi

# Check for project commands
if [ -d ".claude/commands" ]; then
    echo "HAS_PROJECT_COMMANDS=true"
fi

Step 2: Search Appropriate Directories

Based on repo type, search the correct directories:

For marketplace repos:

# Search all plugin command directories + project commands
grep -r "^description:" plugins/*/commands --include="*.md" 2>/dev/null | sort
grep -r "^description:" .claude/commands --include="*.md" 2>/dev/null | sort

For plugin repos:

# Search plugin commands + project commands
grep -r "^description:" ./commands --include="*.md" 2>/dev/null | sort
grep -r "^description:" .claude/commands --include="*.md" 2>/dev/null | sort

For standard repos:

# Search project commands only
grep -r "^description:" .claude/commands --include="*.md" 2>/dev/null | sort

Step 3: Format Output

Parse the grep output and format as:

### Plugin Commands ({plugin-name})

### **/command-name**
Description from frontmatter

---

### Project Commands

### **/command-name**
Description from frontmatter

---

Output Requirements

• Group commands by source (plugin vs project)
• Use bold markdown for command names (### /name)
• Add horizontal rule separator (---) between commands
• Keep descriptions concise
• Show total count at the end by source
• Sort commands alphabetically within each group

When Subagent Pattern Makes Sense

For larger operations (50+ files, complex per-file analysis), consider:

1. One agent to discover all files
2. Divide into groups of ~10-15 files
3. Parallel subagents to process each group
4. Main agent aggregates results

This pattern adds ~2-3s overhead per subagent, so only use when per-item processing time justifies it.