Inspect a project and its results

Use this read-only route when an agent needs to explain an existing project before changing it. Start with a connected Look only key and the first-read walkthrough.

Read the current project

Choose the intended project from list_projects, then call read_project_graph with its returned URL or ID. Keep the returned revision. Read node types, titles, fields, links, result status and any scriptwriter report.

Studio-owned step instructions can be redacted even when their node or trace row is visible. An ordinary API key does not expose that protected text; use the supported result or action instead.

Do not infer a node’s job from its custom title alone. Look up its type in describe_studio or the node reference. The graph returns actual port names; use those names in later edits.

Follow one scene

Take a scene address from the graph’s scene_in or scene_out fields:

{
  "name": "describe_scene",
  "arguments": {
    "project_id": "<url or id returned by list_projects>",
    "scene_id": "<scene_out address returned by read_project_graph>"
  }
}

Compare the original scene, recorded rewrites and deliveries. A rewrite may be the current text even though an older script artifact contains different wording. If the address is no longer carried by the graph, read a fresh graph: regenerating the script can create new scene addresses.

Answer the right historical question

Question	Read
What is currently saved on the canvas?	read_project_graph
Which nodes ran, reused results or failed?	read_run_history
Which inputs produced this generation?	trace, narrowed to its node or artifact
What happened to this scene along the graph?	describe_scene
Where is the completed material?	list_artifacts, then get_artifact

An empty trace does not prove that a node never ran: some operations do not record generation provenance. A visible done result does not prove freshness after input changes. Read text and results before diagnosing a mismatch.

Check explicit shortened/omitted flags in detailed reports. The full node result on the canvas remains the place to review content that does not fit in a report.

Last checked: 15 October 2026. These reads do not generate content or change the graph.