dbx/skills/dbx/SKILL.md

name, version, description, metadata

name	version	description	metadata
dbx	1.0.0	DBX CLI for database schema exploration and read-only queries. When the user needs to list connections, explore tables, describe schemas, run queries, or generate AI-friendly schema context from DBX-managed databases. Do NOT use for write operations unless the user explicitly confirms with --allow-writes.	requires cliHelp bins dbx dbx --help

DBX CLI

Prerequisite: DBX Desktop must be installed and configured with at least one connection. Run dbx doctor to verify setup. Install CLI: npm install -g @dbx-app/cli

Core Concepts

• Connection: A named database connection configured in DBX Desktop (e.g. "prod", "local"). Identified by name.
• Schema: Tables and views within a connection. Listed via dbx schema list.
• Query: Read-only SQL executed against a connection. Gated by --allow-writes and --allow-dangerous-sql.
• Context: Compact schema dump optimized for AI prompts — smaller and more focused than full schema output.

Resource Relationships

DBX Desktop
├── Connection (named)
│   ├── Schema (tables, views)
│   │   └── Table
│   │       └── Column (name, type, nullable, default)
│   └── Query (read-only by default)
└── Context (prompt-optimized schema dump)

Commands

⛔ NEVER bypass dbx CLI for database operations. If dbx returns SQL_BLOCKED or any error, do NOT use Python sqlite3, shell redirects, or any other tool to access the database directly. Always respect the CLI's safety gates. When in doubt, tell the user what the CLI returned and ask for their decision.

1. Check Setup

dbx doctor
dbx capabilities

Use doctor when the agent starts or when a command fails — it reveals whether the desktop bridge, connection DB, and native SQLite loader are available. Use capabilities to check which databases support direct execution vs require the desktop bridge.

2. List Connections

dbx connections list --json

Returns connections without exposing secrets. Parse JSON to present a clean list to the user. Always run this first before any schema or query operation — the user might not remember connection names.

3. Explore Schema

# List all tables in a connection
dbx schema list <connection> --json

# Describe a specific table
dbx schema describe <connection> <table> --json

Use schema list to survey what's available, then schema describe on specific tables the user asks about. Present column names, types, and nullability clearly.

4. Execute Queries

# Read-only query (default)
dbx query <connection> "SELECT ..." --json

# From file
dbx query <connection> --file ./query.sql --json

# With row limit and timeout
dbx query <connection> "SELECT ..." --limit 50 --timeout 10s --json

CRITICAL — Read-only by default. Write operations (INSERT/UPDATE/DELETE) require --allow-writes. Dangerous SQL (DROP/TRUNCATE/ALTER) requires BOTH --allow-writes AND --allow-dangerous-sql. Never add these flags unless the user explicitly confirms a write operation.

If the SQL starts with a dash, separate with --:

dbx query local --json -- "-- comment
select 1"

5. Generate Context for Prompts

# Full schema context
dbx context <connection>

# Filtered to specific tables
dbx context <connection> --tables users,orders,products

Use context when the user wants to write a query but needs schema reference first. Pipe the output directly into the prompt — it's designed for this. Prefer --tables to limit scope and save tokens.

6. Default Connection

Set DBX_CONNECTION to skip the connection argument:

export DBX_CONNECTION=prod
dbx query "SELECT 1" --json
dbx context --tables users

Detect and use this if set in the environment.

Output

Flag	Use Case
--json	Machine-readable, auto-parsed (always use this)
--format csv	Piping to other CLI tools

Errors go to stderr with non-zero exit code. Run dbx doctor first if any command fails unexpectedly.

Error Codes

Code	Meaning	Agent Response
CONNECTION_NOT_FOUND	Connection name doesn't exist	List available connections with dbx connections list --json
SQL_BLOCKED	Write operation attempted without --allow-writes	Ask user: "This is a write operation. Confirm?" Never add write flags automatically.
DBX_NOT_RUNNING	Desktop bridge unavailable	Tell user to open DBX Desktop. Check which commands work without bridge via dbx capabilities.
INVALID_OPTION	Wrong flag or flag value	Check dbx --help and retry
ERROR	Unexpected runtime failure	Run dbx doctor, check logs, retry once

Direct vs Bridge Execution

PostgreSQL, MySQL (and compatible: Doris, StarRocks), SQLite run directly without DBX Desktop. Other database types require the desktop bridge. Check with dbx capabilities to confirm.

Common Pitfalls

1. Wrong connection name — Always list connections first with dbx connections list --json before running schema or query commands. Never assume connection names from conversation context.

2. Schema confusion from context pollution — When the user asks about a table, verify the connection and table exist before running queries. dbx schema list <conn> --json is your verification step.

3. Write operations by accident — Never add --allow-writes or --allow-dangerous-sql unless the user explicitly confirms. When in doubt, ask.

4. Missing desktop bridge — If dbx open or bridge-required connections fail, run dbx doctor and tell the user to open DBX Desktop. Commands that don't require the bridge: connections list, schema list, schema describe, query, context (for PostgreSQL/MySQL/SQLite).

5. Timeout on large queries — Always use --limit 50 --timeout 10s for exploratory queries. Remove or increase limits only when the user explicitly asks for full results.

6. JSON parse errors — Old DBX versions may not support --json on some commands. If JSON output looks malformed, try without --json and parse the human-readable output instead.

Multi-Step Workflows

Explore then Query

1. dbx connections list --json — verify connection exists
2. dbx schema list <conn> --json — survey available tables
3. dbx schema describe <conn> <table> --json — understand target table
4. dbx query <conn> "SELECT ..." --limit 50 --timeout 10s --json — execute
5. Present results to user with row count

Generate Context then Help Write Query

1. dbx context <conn> --tables a,b — get compact schema
2. Read the output, understand relationships
3. Draft the SQL, show it to user for review
4. dbx query <conn> "polished sql" --json — execute after approval

Cross-Connection Comparison

1. dbx connections list --json — identify source and target
2. dbx schema describe <source_conn> <table> --json — get source structure
3. dbx schema describe <target_conn> <table> --json — get target structure
4. Compare and report differences