The Thirteen Skills With Workflow Below 3
Workflow is the weakest dimension in the new batch, with a mean of 5.30. The average hides a sharper fact: thirteen skills score below 3.0, and they cluster by type.
The thirteen:
The shared shape
The tool wrappers — Asana, ClickUp, Calendar, Sheets, Slides — list operations and leave the sequence implicit. They tell the agent what the API can do, not what to do first, what to verify, or what to do when a step fails.
The marketing files — the e-commerce how-to pages — describe a topic and give general advice. They do not ask for the store's actual data, branch on the situation, or define a finished output.
Both groups fail for the same reason: there is no procedure to score. The file is reference material.
What would raise them
A workflow section needs five things:
- Preconditions. What must be known or fetched before the first action.
- Ordered steps. A sequence with a clear starting point and stopping point.
- Inputs and outputs. What each step consumes and produces.
- Verification. How the agent checks that a step succeeded.
- Failure handling. What to do when a call fails, a value is missing, or the user's request is ambiguous.
The best files in the batch do all five. Jira scores 9.5 on workflow because it starts with capability discovery, routes between backends, and gates mutations. k12-lesson-plan-creation scores 9.0 because the lesson plan has timed segments and a defined artifact.
The gap between 1.0 and 9.5 is not domain knowledge. It is the decision to write a procedure.