Skip to Content
Agent connectionsAgent workflows and examples

Agent workflows and examples

A reliable agent workflow reads current records, preserves brand facts, saves with revision checks and inspects rendered artwork.

Plan without changing designs

Find my travel project and read its brand facts. Plan five post ideas using only those facts. Identify missing details. Do not create designs or publish anything.

The agent should discover the project rather than guess its ID. It can read included sources and references to understand the brief.

Create editable posts

Choose three suitable templates from this project. Create a campaign using these approved headlines. Keep the layouts and logo, adapt the images from this project’s library, and render every saved post for review.

Each intended copy uses its own request identity. Retries reuse that identity. The original templates stay independent from their campaign copies.

Edit a saved design

The agent reads the latest design, including its revision, then sends only the intended changes. A supplied props, fields or source value replaces that entire value; omitted fields stay unchanged. The agent must preserve unrelated content when preparing the replacement.

Saving requires expectedRevision. A conflict means another edit happened, so the agent should reread and reconcile. After saving, it renders that exact revision and checks the image.

Review before publishing

Review these saved posts for Instagram. Show the artwork, captions, selected account and proposed schedule. Wait for my instruction before submitting.

Publishing needs both the connector’s Publish permission and your instruction. A future schedule uses an ISO 8601 timestamp with Z or an explicit offset. Example: 2027-02-15T10:00:00+02:00.

If a request returns an unknown outcome, check provider status before retrying. Keep the same request IDs for a retry of the same intended operation. submitted means accepted by the publishing service, not confirmed live.

Boundaries

The connector can return images directly to the agent. It does not provide your browser cookie, cloud account token or a shell. Resize and size-variant generation require Doodi’s design engine to be connected; an agent can also author supported editable React source itself.