Agent Skills#

edg ships with a set of Agent Skills; slash commands that help you generate, validate, debug, and migrate workload configs without leaving your terminal.

Install#

Install all edg skills into your project with one command:

npx skills add codingconcepts/edg --agent claude-code --skill '*' --yes

Or install a specific skill:

npx skills add codingconcepts/edg --agent claude-code --skill edg-config --yes

If you’ve cloned edg and are running Claude Code from the repo root, the skills are available automatically.

Windows users: Git symlinks require Developer Mode enabled or git config core.symlinks true set before cloning. Without this, the symlinks in .claude/skills/ will be checked out as plain text files. Run git clone -c core.symlinks=true to avoid this.

Available Skills#

/edg-config - Generate a workload config#

Generates a complete edg YAML config from a natural language description of your schema and workload. Includes database-specific patterns for all supported drivers.

Example prompts:

/edg-config

I have a users table and an orders table on CockroachDB.
Users have an email, name, and created_at timestamp.
Orders reference a user and have a total, status, and created_at.
I want a 70/30 read/write mix benchmarking order lookups and new order inserts.
Seed with 10k users and 50k orders.

/edg-config

MySQL e-commerce schema: products (name, category, price, stock),
customers (email, name), and a cart_items join table.
Write-heavy workload simulating add-to-cart with Zipfian product selection
(some products are much more popular). Use batch inserts for seeding.

/edg-config

Generate a sync config pair for CockroachDB (pgx) and MySQL.
Tables: users (id, email, name) and orders (id, user_id, amount, status).
1000 users and 5000 orders, batched 100 at a time.
I'll use edg sync run with --rng-seed for deterministic dual-write testing.

/edg-config

MongoDB workload: customers collection with email and name fields,
and an accounts collection referencing customers with a balance field.
Seed 500 customers and 500 accounts, then run random balance lookups.

/edg-config

Cassandra IoT schema: create a keyspace with a devices table (id UUID, name TEXT)
and a readings table (device_id UUID, ts TIMESTAMP, value DOUBLE).
Seed 100 devices and 10k readings, then run time-range queries.

/edg-config

CockroachDB bank workload with staged execution: ramp up with 1 worker
doing mostly reads (90%), then steady state with 10 workers doing heavy
transfers (45%), then cool down with 2 workers falling back to default weights.
Use per-stage run_weights to shift the workload mix at each phase.

What it produces:

• A full YAML config with globals, stages, up, seed, init, run, run_weights, workers, deseed, and down sections
• Appropriate expressions for data generation (gen(), uuid_v7(), distributions)
• Reference data setup via init queries for use in run with ref_rand(), ref_same(), etc.
• Driver-specific SQL patterns (batch expansion, DDL safety, upsert syntax)
• Sync config pairs for cross-database consistency testing - matched schemas with explicit IDs, identical seed args, and driver-specific batch SQL for use with edg sync run
• Uses shorthand features where appropriate: inline args (${expr}), __columns__/__values__ expansion, query type inference, and named templates to reduce boilerplate

For a quick starting point without an agent, edg scaffold generates a config interactively from the command line. See Scaffold.

/edg-validate - Validate and fix a config#

Runs edg validate on a config file, interprets errors, and suggests fixes.

Example prompts:

/edg-validate

Check my config at workloads/bench.yaml for the pgx driver.

/edg-validate

Validate _examples/ecommerce/mysql.yaml and fix any issues you find.

What it does:

1. Detects the driver from the config content (or uses one you specify)
2. Runs edg validate --driver <driver> --config <path> --explain (errors now include YAML line numbers and enhanced explanations)
3. Explains each error in plain language
4. Suggests (or applies) fixes for common issues:
   • Unknown functions or typos in expressions
   • Missing init datasets referenced by ref_* calls
   • Wrong query type (exec vs query) - note that type: is now optional and auto-inferred from the query text
   • Missing count/size on batch operations

/edg-expression - Compose and debug expressions#

Helps you find the right edg expression for a use case, explains how functions work, and debugs syntax errors.

Example prompts:

/edg-expression

I need to generate a price between $1 and $500 with a log-normal distribution
skewed toward cheaper items, rounded to 2 decimal places.

/edg-expression

How do I make two columns mutually exclusive (either an email OR a phone
number, but never both?)

/edg-expression

What's the difference between ref_rand, ref_same, and ref_perm?
When would I use each one?

What it covers:

• All 100+ built-in functions with signatures and descriptions (also available via edg functions [search])
• Distribution selection guidance (uniform vs normal vs zipf vs exponential)
• Reference data patterns (ref_rand vs ref_same vs ref_perm vs ref_diff)
• Dependent column patterns with arg(), cond(), and bool()
• Inline args (${expr}) and __columns__/__values__ shorthand
• The full expr-lang feature set (array functions, string ops, math, conditionals)
• Tips for debugging with edg repl (with tab completion for function names)

/edg-migrate - Convert a config between drivers#

Converts an edg config written for one database driver to another, handling all SQL dialect differences.

Example prompts:

/edg-migrate

Convert _examples/ecommerce/crdb.yaml from pgx to mysql.

/edg-migrate

I have a workload config for PostgreSQL and need it to work on SQL Server (mssql).
The config is at workloads/bench.yaml.

/edg-migrate

Convert my PostgreSQL workload at workloads/users.yaml to MongoDB.

What it translates:

Expression args (gen(), ref_rand(), zipf(), etc.) are driver-agnostic and remain unchanged.

Tips#

• Combine skills: Generate a config with /edg-config, then validate it with /edg-validate, then port it to another driver with /edg-migrate.
• Start with scaffold: Run edg scaffold for a quick interactive starting point, then refine with /edg-config for more complex needs.
• Use the REPL: All expression skills recommend edg repl for interactive testing with tab completion; no database connection needed.
• Discover functions: Run edg functions [search] to find available functions by name (e.g. edg functions ref, edg functions seq).
• Validate after migration: Always run edg validate --driver <driver> --config <path> --explain after generating or migrating a config to catch issues before hitting the database.