Skip to main content

Tool Commands

Custom tools let you extend Agent Actions with deterministic logic. They handle tasks that LLMs can't do alone, like calling APIs, validating data, or transforming outputs.

Commands for discovering and validating tools:

list-udfs​

What tools are available in my project? This command scans your code directory for Python functions decorated with @udf_tool.

agac list-udfs -u <user-code-path> [options]

Discovers all tools and displays their metadata - location, file path, and documentation.

Options:

OptionDescription
-u, --user-codePath to user code directory containing tools (required)
--jsonOutput as JSON for programmatic use
--verboseShow full signatures and docstrings

Examples:

# List tools in table format
agac list-udfs -u user_code/

# Output as JSON
agac list-udfs -u user_code/ --json

# Show full details (signatures, docstrings)
agac list-udfs -u user_code/ --verbose
tip

Use this command to verify which tools were discovered from your code directory before running your agentic workflow.

validate-udfs​

Will my agentic workflow find all the tools it needs? Validates that every impl reference in your configuration points to a real, properly decorated function.

agac validate-udfs -a <agentic-workflow> -u <user-code-path> [options]

Catches misspelled function names or missing @udf_tool decorators before execution begins.

Options:

OptionDescription
-a, --agentAgentic workflow name (required)
-u, --user-codePath to user code directory containing tools (required)

What it validates:

  • All impl references exist in the tool registry
  • No duplicate function names across files
  • All Python files can be imported without errors
  • Functions are properly decorated with @udf_tool

Examples:

# Validate agentic workflow config references
agac validate-udfs -a my_workflow -u user_code/
When to Use

Run this command before deploying agentic workflows to catch tool reference errors early. Ideal for CI/CD pipelines where you want to fail fast on configuration errors.

Limitation

Validates that tool references exist and are properly decorated, but doesn't execute the functions. Runtime errors inside your tool code (like API failures or type mismatches) will only surface during actual execution.

See Also​