agentgraph inspect opens a run’s SQLite checkpoint and artifact store and prints the status of every node alongside a preview of its output. Use it to verify what each node produced, debug a failed run, or read the full artifact from a specific node.
Arguments and flags
The run ID to inspect. Run IDs are printed at the start of every
agentgraph run invocation and listed by agentgraph list.Directory containing run subdirectories. Override if you used a custom
--runs-dir when running the DAG.Show the artifact for a single node instead of all nodes. Combine with
--full to print the complete output.Print the complete artifact content without truncation. By default, long artifacts are truncated at
--truncate characters.Character limit for artifact previews when
--full is not set. Truncated content shows a count of remaining characters and a reminder to use --full.Examples
Output format
inspect prints two sections: a run info panel followed by a nodes table.
Run info panel — shows run ID, DAG name, start time, finish time (or a note if the run is still in progress), and the input values used.
Nodes table — one row per node, with four columns:
| Column | Contents |
|---|---|
| node | Node ID as declared in the DAG |
| status | Current status (see values below) |
| artifact | First 12 characters of the SHA-256 digest, or the error message for failed nodes |
| updated | Timestamp of the last status change |
inspect prints an artifact preview panel for each completed or skipped node. For failed nodes, the error message appears in the table row.
Node status values
| Status | Meaning |
|---|---|
completed | Node finished successfully and wrote an artifact |
failed | Node encountered an error; the error message appears in the artifact column |
running | Node is currently executing (live run only) |
pending | Node is waiting for its dependencies to complete |
awaiting_approval | Node is an approval_gate paused for human review |
skipped | Node was skipped due to a when: condition evaluating to false |
If
finished shows — (running or interrupted), the run did not complete cleanly. Use agentgraph resume or re-run with --run-id to continue from where it stopped.