Passed 18 of 22 test cases with the skill, and 12 of 22 without it.
Without skill55%
With skill82%
Checks it passed that the model alone missed
Explains the rationale for each tab and section grouping.
Presents the layout with Tab Break and Section Break formatting.
Provides an explanation for why each tab and section group exists.
and 6 more in the full report
6 fixed12 passed either way4 still failing0 broke
Pass-rate lift
+27 pp
Tokens
+17%
Agent turns
0%
Response time
+0%
Model gemini-3.6-flashMeasured 8/15/2026v118 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 6
Prompt
I need a complete Frappe DocType design for managing Vendor Onboarding in our ERPNext setup. We want to collect basic company information, primary contact person, bank/payment details, tax registration numbers, and track their approval workflow status. Can you propose the field layout, form structure, and field names, along with explanations for how the layout is organized?
What was checked, without and with the skill
Checked
Without skill
With skill
Identifies the DocType purpose, primary user, business lifecycle, and DocType category before presenting the field schema.
Missed
Passed
Organizes fields into explicitly named Tab Breaks and Section Breaks.
Passed
Passed
Provides an explanation for why each tab and section group exists.
Missed
Passed
case-01 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md34 linessingle file
Design every DocType as a human-facing form, not a flat database table.
Before adding fields, identify:
the DocType's purpose and primary user
the real business process or lifecycle it supports
the relationships, status, reporting, and search needs
whether it is master data, a transaction, a child table, or a configuration DocType
Field discipline:
Add only fields that are required for identity, workflow, relationships, decisions, audit, reporting, or search.
Do not add speculative, vanity, duplicate, or generic fields unless the use case clearly needs them.
Make required fields truly required, not merely convenient.
Use child tables only for real one-to-many repeating data.
Prefer human-readable labels and stable snake_case fieldnames.
Form layout:
Do not dump all fields into one tab or one section.
Use Tab Break for major mental models only when the form has enough fields to justify tabs.
Use Section Break for logical groups inside a tab.
Use Column Break to create balanced, scan-friendly rows.
Put the title, status, primary links, and most common required inputs first.
Put advanced, rare, admin, or integration fields later, and fold them when appropriate.
Avoid empty tabs, thin sections, and layout breaks that exist only for decoration.
Useful tab patterns include:
Overview for identity, status, and primary summary fields
Details for the main business data
People or Parties for owners, customers, suppliers, employees, or contacts
Schedule for dates, timelines, reminders, or recurrence
Finance for amounts, accounts, taxes, or pricing
Settings for infrequent configuration
Activity or Audit for comments, references, or system-facing traceability
When proposing or creating a DocType, present the layout grouped by tab and section, and explain why each group exists. If a requested field does not belong, call it out and omit it unless the user confirms the need.