silver-evaluate
Define an evaluation question and method, prepare tasks, inspect a visualization or prototype, capture sanitized observations, and produce findings and recommendations. Use for usability testing, expert review, or feedback analysis distinct from deterministic conformance.
- Category
- Testing and evaluation
- Status
- Experimental
- Use when
- plan a usability test
- run an expert review of this prototype
- evaluate this design and separate observation from interpretation
- produce evidence-linked evaluation findings
- distinguish user-tested findings from automated conformance failures
- Works on
- a sketch, visualization, prototype, design specification, or map to evaluate
- an evaluation question, method, and participant/reviewer set
- Produces
- evidence-linked findings and recommendations that keep observed behavior, interpretation, confidence, and deterministic-conformance results distinguishable
- Requires
- node
- Works with
- Claude Code, Codex, Cursor, OpenCode, Amp, Gemini CLI, GitHub Copilot, VS Code, Claude.ai
- Tags
- usability-testing, heuristic-evaluation, design-critique, expert-review, severity
- Source
- thejparsenault/silver-design-framework by JP Arsenault · MIT
- Related skills
- silver-ideatesilver-measureinterface-review
Instructions from SKILL.md
Evaluate design
Workflow
- Define the decision, question, method, participants or reviewers, and tasks.
- Inspect the pinned visualization or prototype and capture only sanitized observations.
- Separate observed behavior from interpretation and deterministic check findings.
- Produce evidence-linked findings and recommendations for explicit acceptance.
Run the guarded file operation through the CLI when durable outputs are ready:
.silver/bin/silver invoke --scaffold evaluate .
.silver/bin/silver invoke evaluate <request.json> .The scaffold prefills timestamps, provenance, pinned context, required checks, and
expected_integrity. Replace every silver-scaffold-placeholder before invoking;
the CLI refuses a request that still contains one.
Done
- Satisfy: question-declared, observations-sanitized, interpretation-separated, conformance-distinct.
- Evaluate quality: Observations, interpretations, confidence, and recommendations remain distinguishable and traceable.
- Emit a valid
silver/skill-result/v2record separating execution, acceptance, and downstream readiness. - Recommend follow-up skills; never start them automatically.
Boundaries
- Do not imply participants or sessions existed when performing only a planned or expert review.
- Do not turn conformance failures into fabricated user evidence.
Supporting files
4 more files in this skill's folder. Markdown opens as a page, other text files open as source, and anything else downloads as it is.