The outcome

Share an inspectable automation recipe that behaves predictably across users and environments.

goose is a open-source general-purpose AI agent for desktop, terminal, and API workflows. Running local coding and knowledge-work agents with selectable model providers, extensible MCP tools, reusable recipes, and configurable security controls. This guide narrows that broad capability into one repeatable outcome, with checkpoints that keep the source material and your judgment in the loop.

Before you begin

Set the boundary before the tool starts.

Choose one real task, identify who will use the result, and decide what evidence or test will make the result acceptable. Gather only the source material needed for that task. If the work contains confidential, personal, regulated, or client-owned information, confirm that the platform and account are approved before sharing it.

Troiana principle

AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.

Step by step

A workflow you can repeat.

  1. 01

    Select a stable manual workflow and map its inputs, outputs, dependencies, extensions, credentials, approval points, retries, and terminal states.

  2. 02

    Create a recipe with typed or clearly documented parameters, narrow instructions, pinned extensions, bounded steps, and no embedded secrets.

  3. 03

    Run it in a disposable environment with normal, missing, malformed, adversarial, tool-failure, timeout, and duplicate-execution cases.

  4. 04

    Compare results with the manual baseline, inspect every side effect and model-dependent decision, and add deterministic checks where possible.

  5. 05

    Version and review the recipe as code, document ownership and rollback, and distribute it only with required permissions and provider assumptions.

Working standard

What good use looks like.

  • Never embed credentials in recipes.
  • Pin and review extensions.
  • Test retries and duplicate execution.

A local agent can still expose data or damage files through model requests, shell access, extensions, browsers, and remote APIs. Use sandbox and permission controls, distrust instructions embedded in files or web pages, install extensions and skills only from reviewed sources, scope credentials and directories narrowly, pin reusable recipes, require approval for side effects, and verify all outputs and changes independently.

Official references

Check the current product documentation.

Features, plan limits, availability, and data controls change. These official pages are the starting points used for this collection.