| name | design-rubric |
| type | skill |
| category | instruction |
| description | Design-stage fitness rubric — persona immersion, scenario design, dimensions that define what excellence looks like for the people a feature serves. Two modes — author (produce a rubric for a new spec) and critique (red-team an existing spec). Output lives on the spec, not in the verification brief. Owned by pauli. |
| triggers | ["design the rubric","fitness criteria","what does success look like","evaluate design","red team this spec","will this overwhelm the user","AC for fitness"] |
| modifies_files | true |
| needs_task | false |
| mode | conversational |
| domain | ["quality-assurance","design"] |
| version | 0.1.0 |
| permalink | skills-design-rubric |
/design-rubric — Design-Stage Fitness Rubric
Create or red-team a qualitative Fitness Rubric for user-facing features to define what excellence looks like. The rubric lives directly on the spec or epic body under ## Fitness Rubric.
Authoring or red-teaming a rubric is premise-and-fit judgment, not template-filling: it runs as pauli (plugins/pkb/agents/pauli.md).
Qualitative Bar Signals
A rubric is required if:
- Criteria use qualitative terms ("intuitive", "calm", "beautiful").
- The brief describes emotional/cognitive user states ("anxious", "depleted").
- The primary consumer is a human.
- Two evaluators could disagree on PASS/FAIL using the same evidence.
- The task fits a fitness-for-purpose threshold.
If these apply and a rubric is missing at verify time, halt with: REVISE — fitness rubric missing; escalate to pauli/design-rubric.
The Four-Step Rubric
- Persona: One situational paragraph detailing the user's emotional state, cognitive constraints, and what success feels like. Do not write generic demographic profiles.
- Three Scenarios: One short paragraph each.
- Golden: Primary happy path.
- Stressed: User under pressure (tired, late, divided attention).
- Edge: Incomplete, late, or corrupted data.
- Dimensions: 3 to 5 qualitative judgment questions. Focus on comprehension, cognitive load, task fitness, and degradation.
- Quality Spectrum: Narrative description (not numeric) of excellent vs. poor execution for each dimension. Mark one critical dimension as
**load-bearing** if applicable (marsha cannot PASS if this dimension is poor).
Self-Instance Requirement
When the feature surfaces the user's OWN data or view ("show me my sessions", "my tasks", "where I was"), a generic instance of the thing appearing is not proof. The rubric must name the principal's concrete identifying signal — the exact fields, account, host, or launch-context that mark the artifact as HIS instance — so verification can later reproduce the principal's literal view and confirm his own data is present. Treat this as load-bearing on any "show MY X" feature: a generic-instance pass is a FAIL.
Output Schema
Save directly to the spec or epic task body:
## Fitness Rubric
Golden:
Stressed: ...
Edge: ...
— excellent: ; poor:
