> ## Documentation Index
> Fetch the complete documentation index at: https://agents.nanonets.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Get Task Details

> Reads the task that started this one through a hook: its conversation and every step's tool call and result.

Reads everything the task that started this one through a hook did: its conversation and every step's tool, status, error and result. Display name **"Get Task Details"**. Off by default.

Use it in the agent a hook targets, to build a structured summary of the task that fired the hook. The source agent does not need to produce consistent JSON: the reading agent works from the conversation and tool results and shapes the payload itself. The task's state, when its agent has one, is already in the hook-started task's first message.

## Authentication and enablement

No integration. Off by default; enable it on the agent's tools list. Read-only. It works only in a task a hook started.

## Inputs

| Field | Type | Required | Description |
| - | - | - | - |
| `step_id` | string (UUID) | no | A `step_id` from the overview: returns that step's full result instead of the overview. |
| `offset` | integer | no | With `step_id`: where to continue a long result, in bytes (the previous page's `next_offset`). |

Read it in two passes: call it with no arguments for the overview, then call it with the `step_id` of each step whose full result you need.

## Which task it reads

Only the source task of the hook that started the running task, as the platform recorded it when the hook fired (the hook's run record, which nothing can edit), in the running task's workspace. There is no way to name another task.

## Output

Overview (no `step_id`):

```json theme={null}
{
  "task": { "id": "d5b3…", "agent_group_id": "2639…", "title": "05 2026 Payment Reconciliation", "status": "waiting_for_input", "source": "trigger", "created_at": "…" },
  "messages": [
    { "role": "user", "text": "Please run May payroll", "file_names": ["05_2026_Payment_reconciliation.xlsx"], "file_ids": ["8e20…"], "created_at": "…" }
  ],
  "steps": [
    { "step_id": "c533…", "tool": "reconciliation_matcher", "status": "completed", "args_preview": "{…}", "result_preview": "{\"lines\":[…", "result_bytes": 75900, "created_at": "…" },
    { "step_id": "9f58…", "tool": "composio_gmail_send_email", "status": "failed", "error": "…", "result_bytes": 0, "created_at": "…" }
  ]
}
```

* `messages`: each turn's text, file names and file ids, oldest first. Read a file's contents with `read_file` and its id. A message over 4,000 bytes is cut and marked `truncated`.
* `steps`: every step except internal bookkeeping, oldest first, with a 300-byte preview of its arguments and a 1,200-byte preview of its result. `result_bytes` is the full result's size.
* The overview fits one budget (120 KB): past it, the oldest steps and then the oldest messages are left out and counted in `older_steps_not_shown` / `older_messages_not_shown`. At most the latest 120 steps and 40 messages are shown.

One step (`step_id`):

```json theme={null}
{ "task_id": "d5b3…", "step_id": "c533…", "tool": "reconciliation_matcher", "status": "completed",
  "offset": 0, "total_bytes": 75900, "next_offset": 50000, "result": "{\"lines\":[…" }
```

A result is what the agent that ran the step saw of it: the same projection the decision loop applies, without generated files' bytes, full lookup catalogs or keys a tool marks feed-only. Values are passed through exactly as stored. It comes in pages of 50,000 bytes; `next_offset` is set while more remains, and an offset inside a character starts at the next one.

## Limits and side effects

* No side effects, no model call, no network.
* One task only: the source of the hook that started the running task.
* Message file URLs are left out (use the file ids). A step's result can still carry the links its tool returned.

## Expected errors

* `this task was not started by a hook`: called in a task no hook started.
* `step <id> is not a step of the source task`: the step belongs to another task.
* `step_id must be a step UUID`: the input is not a UUID.
* `offset … is past the end of the result`, or `offset pages a step's result`: an offset without a `step_id`.
* `no workspace in context`: called outside a running task.


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