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

# Writing good tasks

> Give a goal, not a procedure. The agent already knows your brand, so spend the description on what's specific to this job.

A task description has one job: say what the finished work should be, and what it must respect. Brand background, audience profiles and voice rules don't belong in it, because the agent already reads them.

## Describe the outcome and let the agent plan the steps

The agent plans its own steps, so describe the finished result and the limits it must respect. A required SOP is part of the job's input, so attach it instead of recreating its steps.

| Instead of              | Write                                                                                                                         |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
| Audit our pricing page. | Audit `/pricing` for SEO issues affecting organic sign-ups. Use the attached keyword CSV and prioritize this quarter's fixes. |

## Your Brand Kit already carries the standing context

The agent reads your [Brand Kit](/agents/set-up-brand-kit) on every task. It already knows who you are, your audience, your voice, and your authors. It also reads the rules you've set for each content type. Don't copy any of that into the task.

If a rule should apply to every future task, change it in the Brand Kit.

## Put this job's specifics in the description

The task description is for **job-specific context**. Name the exact page or keyword set, the deadline, and any constraint that changes the answer.

Include the audience for this one piece when it differs from your usual audience.

Attach source material the agent must use, such as a data CSV or an SOP.

Anything with an exact answer is better [tagged than described](/agents/tagging-context). A playbook, an author, an audience or a knowledge base source goes in as a reference rather than a phrase the agent has to interpret.

## Comments preserve the work and its context

When a deliverable misses the mark, leave feedback in its comments rather than rejecting it. Say what to redo, what to drop, and what to prioritise. The agent revises while keeping the context it already gathered. Rewriting the task from scratch throws that away.

## What's next

<CardGroup cols={2}>
  <Card title="Common tasks" href="/agents/common-tasks" icon="book-open">
    Copyable examples for common marketing work.
  </Card>

  <Card title="Approvals" href="/agents/approvals" icon="circle-check">
    See how much you review.
  </Card>
</CardGroup>
