| name | aget-create-rubric |
| description | Create scoring rubrics with domain-adaptive structure, extreme anchoring, and governed storage. ADR-008 Generator level for rubric creation. |
/aget-create-rubric
Create scoring rubrics with domain-adaptive structure, extreme anchoring, and governed storage. Wraps SOP_RUBRIC_CREATION.md and RUBRIC.template.md into a Generator-level skill.
Purpose
Six agents actively produce rubrics with four incompatible storage conventions and no shared construction methodology (L731). The SOP and template are mature (v1.0, active since 2026-02-13) but require manual discovery and application. This skill is the ADR-008 Generator level for rubric creation, progressing from the current Strict level (SOP + template).
Evidence: L731 (Fleet Rubric Adoption Snapshot), L730 (Domain-Fit Gap), L729 (Assessment Tool Recursion), L518 (Conformance Assessment Pattern). SOP + template mature 4+ months. 6 agents producing rubrics without tooling.
Input
$ARGUMENTS - Description of what the rubric will evaluate
Examples:
/aget-create-rubric skill proposal readiness — Rubric for evaluating skill proposals
/aget-create-rubric CI/CD investment priority — Rubric for CI/CD decisions
/aget-create-rubric session process quality — Rubric for session quality assessment
create a rubric for release readiness — Natural language trigger
Execution
Step 1: Domain Identification
Prompt for or derive from $ARGUMENTS:
- Evaluation target: What will be scored? (artifact, process, decision, behavior)
- Decision this supports: What action does scoring inform? (approve/reject, prioritize, improve)
- Assessor: Who scores? (human, agent, hybrid) — per SOP Phase 1 Step 1 prerequisite (L730)
If assessor is not provided, this is a blocking prerequisite — ask before proceeding.
Step 2: Domain Classification
Classify the rubric domain to determine label set:
| Domain Type | L0 | L1 | L2 | L3 |
|---|
| Conformance | Non-Conformant | Baseline | Compliant | Exemplary |
| Creative Quality | Unrecognizable | Emerging | Achieved | Transcendent |
| Maturity | Initial | Developing | Defined | Optimizing |
| Process Quality | Ad-hoc | Repeatable | Measured | Optimizing |
Present classification and label set for confirmation before proceeding.
Step 3: Dimension Scaffolding
Propose 3-7 dimensions with definitions. For each dimension, ask:
- "Is this pass/fail or graduated?" (constraint vs dimension — L730)
- If pass/fail: move to Constraints section (binary eligibility gates)
- If graduated: keep as scored dimension
Step 4: Extreme Anchoring (Mertler 2001)
For each dimension, write level descriptions in this order:
- L3 first (best case) — What does exemplary look like?
- L0 second (worst case) — What does failure look like?
- L1 and L2 (interpolate) — Fill the middle from the extremes
This is the Mertler construction method. Writing extremes first prevents anchor collapse.
Step 5: Anti-Pattern Check
Verify the rubric does NOT contain:
| Anti-Pattern | Source | Detection |
|---|
| Score multipliers | L018 | Warn if weights sum >100% or any weight >40% |
| Constraints as dimensions | L730 | Any dimension with only 2 meaningful levels = should be a constraint |
| Missing assessor identity | L730 | Blocking prerequisite — must be set in Step 1 |
| Achievement framing | Goodhart | Descriptions should say what behavior looks like, not what it achieves |
Step 6: Fleet Lesson Surfacing
Display relevant rubric design lessons:
- L018: No score multipliers
- L183: Scoring is a tool role
- L679: Rubrics = personalization mechanism
- L705: Cross-agent traceability
Step 7: Quality Checklist (SOP Phase 3)
Run the SOP_RUBRIC_CREATION.md Phase 3 quality checklist automatically:
- All dimensions have L0-L3 descriptions
- Labels match domain type
- Constraints are separated from dimensions
- Scoring procedure is documented
- Assessor is identified
Step 8: Write Rubric
Write to rubrics/RUBRIC_{name}_v1.0.md using the RUBRIC.template.md structure.
Step 9: Update Index
If rubrics/README.md exists, append entry to the rubric table.
Step 10: Report
=== Rubric Created ===
File: rubrics/RUBRIC_{name}_v1.0.md
Domain: {domain_type}
Dimensions: {count}
Constraints: {count}
Assessor: {assessor}
Labels: {L0} / {L1} / {L2} / {L3}
Requirements
| ID | Requirement | Source |
|---|
| REQ-RUB-001 | Create-rubric SHALL identify assessor before scaffolding dimensions | SOP Phase 1 Step 1, L730 |
| REQ-RUB-002 | Create-rubric SHALL classify domain and apply domain-adaptive labels | L730 |
| REQ-RUB-003 | Create-rubric SHALL use extreme anchoring (L3 then L0 then interpolate) | Mertler 2001, SOP Phase 2 |
| REQ-RUB-004 | Create-rubric SHALL separate binary constraints from scored dimensions | L730, CMMI/CLEAR pattern |
| REQ-RUB-005 | Create-rubric SHALL run quality checklist before writing | SOP Phase 3 |
| REQ-RUB-006 | Create-rubric SHALL store output in rubrics/ directory | C092 convention |
| REQ-RUB-007 | Create-rubric SHALL NOT use score multipliers or weights >40% | L018 |
Constraints
- C1: MUST identify assessor before proceeding past Step 1 — this is a blocking prerequisite
- C2: MUST separate pass/fail constraints from graduated dimensions — mixing them inflates or deflates scores
- C3: MUST write L3 and L0 before L1 and L2 — prevents anchor collapse (Mertler 2001)
- C4: MUST write to
rubrics/ directory — no other storage location is governed
Traceability
| Link | Reference |
|---|
| Proposal | SP-003 (planning/skill-proposals/PROPOSAL_aget-create-rubric.md) |
| SOP | SOP_RUBRIC_CREATION.md v1.0 |
| Template | rubrics/RUBRIC.template.md |
| Ontology | ONTOLOGY_scoring_rubric_v1.0.yaml |
| ADR-008 | Advisory -> Strict -> Generator progression (this skill = Generator) |
| L-docs | L729 (Recursion), L730 (Domain-Fit Gap), L731 (Fleet Snapshot), L518 (Conformance), L572 (Ontology), L679 (Personalization), L705 (Cross-Agent) |
| Research | Mertler 2001, Brookhart 2013, Stevens & Levi 2013, CMMI, CLEAR, Galileo |
aget-create-rubric v0.1.0
Category: Governance
ADR-008 Level: Generator (wraps SOP + template)