> ## Documentation Index
> Fetch the complete documentation index at: https://graph.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# plan_loader (workbench)

> Finds an existing plan from a name or description and loads it into the workbench.

`plan_drafter` runs it as a subagent.

| Model role | Runs in | Max iterations |
| - | - | - |
| `chat` | `graph wb` only | default |

## Tools

* `workbench__list_artifacts`
* `workbench__show_artifact`
* `workbench__load_artifact`

## Input

**request** `string` *required*

What the user asked for, in their words, naming or describing the plan

## Output

**loaded** `boolean` *required*

True when exactly one plan matched and it is now the draft

**identifier** `string` *required*

The loaded plan's identifier, or empty when nothing was loaded

**candidates** `string[]` *required*

Identifiers of the plans that could match when the request was ambiguous

**note** `string` *required*

One sentence on what happened, for the agent that asked

## System prompt

> **Show the system prompt**
>
> ```markdown theme={null}
> You load a plan into the graph plan workbench. Every call passes kind: plan. List the catalog with workbench__list_artifacts, read likely matches with workbench__show_artifact when the name alone doesn't settle it, and load the single best match with workbench__load_artifact.
>
> - Load only when one plan clearly matches the request. When several could match, load nothing and return them as candidates.
> - When nothing matches, load nothing and say so in the note.
> - workbench__load_artifact fails when the current draft has unsaved changes. Never pass overwrite_draft; report the failure in the note so the user can decide.
> ```

```bash Customize theme={null}
graph agent show plan_loader > .graph/agents/plan_loader.yaml
```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.