Specimen first
Fidelity and granularity controls for AI coding agents — the judgment a seasoned engineering team would apply, made explicit. Execution control + surface control. Model-agnostic. MIT.
npx -y skills add knowledgeco/airworthiness --skill specimen-firstAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 27 days oldThe repository was created 27 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Explore and compare materially divergent product surface directions on representative high-pressure screens or workflows before extracting tokens and reusable components. Use before creating a new design system or broad UI, when no surface direction has been chosen, when "make it unique" is the only identity guidance, when competing directions need an operator decision, or when an existing product looks like a themed component library.
SKILL.md
3.8 KB, as published. Nobody here has run it
Specimen First
Failure mode
Prevent a generic design system from hardening before the product has demonstrated an identity worth repeating.
Governing principle
Specimen first; system second. Reusable rules must be extracted from a direction that survived representative pressure and an actual operator selection.
Required sequence
- Establish posture. Read local product and surface authority. If posture is missing, draft explicit proposals and keep the operator choice pending. Treat nested
lockedvalues in a machine-proposed constitution as proposed comparison conditions, not operator-approved constraints. - Choose pressure. Select the smallest set of workflows and states that exposes real design tradeoffs: dense work, first run, failure or degradation, complex input, narrow transformation, high-volume data, or consequential confirmation as relevant.
- Make directions diverge structurally. Vary hierarchy, spatial model, navigation, disclosure, workflow sequence, density, and state treatment. A palette, font, illustration, gradient, or radius swap alone is not a direction.
- Render comparable specimens. Give each direction equivalent content, states, viewport pressure, and task goals. Do not let one option win through greater polish or easier sample data.
- Compare consequences. State strengths, costs, risks, what each direction makes easier or harder, and a machine recommendation with honest confidence.
- Return the choice. Leave selection blank until the operator selects, rejects, combines, or declares that no direction is ready. Do not continue merely to keep a pipeline moving.
- Refine selected evidence. Resolve weaknesses in the chosen or combined direction on the same pressure set before broad rollout.
- Extract the system. Derive tokens, component families, interaction and writing patterns, responsive and state rules, and exceptions from successful specimens. Record library defaults retained, modified, replaced, and still dominant.
Stage gates
- Do not extract a product-wide system while direction status is pending or rejected.
- Do not retrofit specimens to a preselected component kit.
- Do not present materially equivalent skins as operator options.
- Do not fabricate selection, taste, approval, or a human quality verdict.
- Do not force a winner;
no_direction_readyis a complete and honest outcome. - Do not replace accessible or conventional utility behavior solely to increase novelty.
Mechanics
- Read specimen-protocol.md to select pressure cases and run stage gates.
- Read direction-comparison-format.md to create an auditable, non-rigged decision handoff.
- Read system-extraction.md only after a valid selection, when deriving reusable rules and auditing library residue.
Optional composition
This skill works independently. If installed, use surface-charter for the governing posture, statewright for deeper state coverage, voicekeeper for copy behavior, surface-genealogy for rejected directions and durable exceptions, and surface-witness for independent convergence review. A decision-handoff skill may carry the operator fork, but its absence never authorizes the agent to choose for the operator.