Build an advisor workflow — advisor

A single worker agent produces output; a higher-tier advisor agent evaluates it and either approves or requests a revision. The worker revises until approved or max_revisions is reached.

Usage

advisor_workflow(worker, advisor, max_revisions = 3L, state_schema = NULL)

Arguments

worker: An Agent object that produces the draft (typically a lower-cost model such as Haiku).
advisor: An Agent object that evaluates the draft (typically a higher-capability model such as Opus).
max_revisions: Integer. Maximum number of revision cycles (default 3L).
state_schema: A WorkflowState or NULL (uses default). The default schema provides: messages (append), latest_draft (overwrite), advisor_feedback (overwrite), advisor_verdict (overwrite), revision_n (overwrite).

Value

A compiled GraphRunner.

Details

Graph: START -> worker -> advisor -> (approved -> END | revise -> worker).

Examples

if (FALSE) { # \dontrun{
runner <- advisor_workflow(
  worker  = agent("writer",  ellmer::chat_anthropic(model = "claude-haiku-4-5-20251001")),
  advisor = agent("advisor", ellmer::chat_anthropic(model = "claude-opus-4-6"))
)
result <- runner$invoke(list(messages = list("Write a concise explanation of R6 classes.")))
result$get("latest_draft")
} # }