agentgraph validate parses your DAG YAML and checks it without executing anything. It resolves the dependency graph into topological waves, reports the declared inputs and outputs, and inspects every Jinja template for variables that are not accounted for. Run it before agentgraph run whenever you edit a DAG file.
Arguments and flags
Path to the DAG YAML file to validate.
Declare an input key so the template-variable checker treats it as known. Accepts
key=value or just key. Repeat the flag for multiple inputs. Without this flag, any template variable not covered by depends_on or the DAG’s own inputs: block is flagged as a warning.Expected output
A valid DAG prints a summary of waves, declared inputs, declared outputs, and any warnings:Examples
Declared inputs table
When the DAG declares aninputs: block, validate prints each key with its type, required/optional status, default value, and description. This is the same information agentgraph run uses to validate --input values at execution time.
| Column | Meaning |
|---|---|
| key | Input name, usable as {{ key }} in node prompts |
| type | Declared type (string, number, etc.) |
| required / default | Whether a value must be supplied, and the fallback if not |
| description | Human-readable description from the DAG file |
Declared outputs table
When the DAG declares anoutputs: block, validate prints each output path and the node whose artifact it maps to. dagraph writes these files after a successful run completes.
Template variable warnings
If a node prompt references a Jinja variable that is neither independs_on, a known reserved loop variable (iteration, previous_output, evaluator_feedback, candidate, index), nor a declared --input, validate prints a warning:
agentgraph run from executing the DAG, but the template will render an empty string for the unknown variable at runtime.