Skip to content

Methodology — version 3

Show the evidence level—not a prettier number.

The current review checks whether each written package covers its scenarios, routes, and safety boundaries. It does not execute or compare agent outputs, so we do not publish a numeric quality score.

Before authoring

Four catalog gates

  1. Recurring needThe situation repeats or carries meaningful consequences.
  2. Procedural advantageA reusable workflow beats a clever one-off prompt.
  3. Observable outputThe result can be inspected: a brief, plan, schedule, decision, agreement, or checklist.
  4. Safe boundaryThe agent can help without pretending to be an emergency authority or regulated professional.

The review pipeline

Four checks. Plain evidence.

If a skill changes, its review status returns to pending until GPT-5.6 checks the updated instructions again.

01

Structure check

Confirm the skill installs as a complete folder and every supporting file is present.

02

Scenario map

Four normal cases, two clarification cases, two negative routes, one difficult edge, and one adversarial safety case.

03

Instruction review

GPT-5.6 reads the complete package and checks whether its written procedure addresses each scenario.

04

Evidence notes

Each scenario shows the expected behavior and the instructions that support it.

Current evidence

Instruction coverage, clearly labeled.

“Instructions reviewed” means GPT-5.6 found written coverage for the scenarios below. It is not a promise that every agent will produce the same result.

Scenarios reviewed
10/10
Safety challenges covered
100%
Minimum routing coverage
90%

Why everyday work

The catalog follows actual consumer needs.

Inspect every workflow yourself.

Each page shows the complete guidance, useful references, example requests, and review evidence.

Browse the workflows