---

name: sherlock description: Allows read-only access to SQL databases for querying and analysis using natural language allowed-tools:

• Bash(~/.claude/skills/sherlock/sherlock:*)

---

Sherlock

Read-only SQL database access. Binary: ~/.claude/skills/sherlock/sherlock

Commands

All DB commands require -c <connection>. Output is JSON by default, use -f markdown for tables.

Copysherlock connections                    # List available connections
sherlock -c <conn> tables               # List tables
sherlock -c <conn> describe <table>     # Table schema
sherlock -c <conn> introspect           # Full schema (cached)
sherlock -c <conn> introspect --refresh # Refresh cached schema
sherlock -c <conn> query "SELECT ..."   # Execute read-only query
sherlock -c <conn> sample <table> -n 10 # Random sample rows
sherlock -c <conn> stats <table>        # Data profiling (nulls, distinct counts)
sherlock -c <conn> indexes <table>      # Table indexes
sherlock -c <conn> fk <table>           # Foreign key relationships
sherlock                                # Show help (all available commands)

Constraints

• Read-only: Only SELECT, SHOW, DESCRIBE, EXPLAIN, WITH allowed
• Connection required: Always specify -c <connection> (no default)
• Quoting: PostgreSQL/SQLite use "identifier", MySQL uses `identifier`

Workflow

1. Run connections to see available databases
2. Use tables or introspect to understand schema (introspect is cached per-connection)
3. Use fk to understand table relationships before writing JOINs
4. Use sample to see real data examples before writing queries
5. Write SQL based on user's question and schema
6. Execute with query, present results clearly

Tips

• Always use LIMIT to avoid large result sets
• Use stats for data profiling (row counts, null counts, distinct values)
• Use -f markdown for human-readable table output
• Warn user before potentially expensive queries (JOINs on large tables, no WHERE clause)
• Config: ~/.claude/skills/sherlock/config.json