Skip to content
ThorbisInterface Guidelines (opens in a new tab)

Linear deep study

Operational grammar, density, hierarchy, and interaction patterns. All research.

Linear, deeper: calm density is an operating philosophy

Linear remains Thorbis's external source of truth for operational grammar. A deeper pass through its design writing, 2026 interface refresh, dashboard research, mobile redesign, agent guidelines, and current product captures revealed several refinements that were only implicit before.

Linear is inherited by every Thorbis rule and specialist skill; it is not a separate skill. This section records evidence, not another source of law. Open the smallest owning skill first, then use the Thorbis × Linear pattern matrix only when the governing operational pattern remains ambiguous. Select one mapping, open its Thorbis owner, and prove the required states.

  • Attention must be earned. Linear deliberately dims navigation and supporting iconography so the work surface wins. Thorbis chrome recedes after orientation; the current job, customer, invoice, schedule, or report owns the strongest contrast.
  • Structure should be felt before it is drawn. Alignment, tone, spacing, grouping, and stable geometry carry hierarchy before separators. A border must communicate a real boundary, not compensate for weak composition.
  • Remove noise before information. Linear preserves rich density while reducing visual competition. Thorbis should not hide useful operational context merely to look minimal.
  • Same action, same neighborhood. Headers, view controls, and repeated actions remain predictable across workflows. Jobs, Money, Customers, Inventory, and Sites may specialize their work without inventing new chrome.
  • Dashboards are decisions, not galleries. A canonical dashboard declares audience, decision, cadence, owner, and freshness. Metrics include enough comparison or expected-range context to change that decision, not just decorate a home screen.
  • AI inhabits the product. Suggestions and agent work use Thorbis primitives and relationships instead of a separate AI costume. Machine suggestion stays distinguishable from committed state.
  • Agent identity and state are explicit. The Resilience skill owns the contract: an agent is visibly non-human, never presented as a person, acknowledges invocation immediately, exposes lifecycle state compactly, and leaves the accountable person or role clear.
  • Automation earns autonomy. Move by category from observe → suggest → approve → automate. Money, destructive work, permissions, customer communication, and public actions keep stronger control.
  • Constraints move downward. If a rule can live in a primitive, schema, generator, tool, lint rule, or verification check, do not leave it as prose an agent must remember.
  • Design tooling is infrastructure. Linear's internal token and feature-comparison tools make refinement cheap. Thorbis should make themes, density, component states, tokens, and before/after comparisons inspectable in the real product.
  • Paper cuts count. Premium quality is cumulative. Repeated small friction is a design defect even when each instance looks too minor for a feature project.

Primary reading: A calmer interface for a product in motion, How we redesigned the Linear UI, Why is quality so rare?, Dashboard best practices, Triage Intelligence, Agent Interaction Guidelines, and How we built Linear Agent. The maintained source index is in the pattern matrix.

Applied bookmark research

Reviewed 2026-09-06. These examples reinforce the existing owners; they do not introduce another visual style or claim the product behavior is implemented. Read the 26-source review and access limits.

Make the task easier, not the screen busier

A feature can look complete while the dispatcher still has to reconstruct context or wait for unrelated content.

Start from one person, interruption, and completed task. Keep the current job and its next action usable while optional evidence arrives. Treat the real constraint as part of the brief, not something polished away afterward.

Thorbis example. For assigning a tech, show the job, relevant availability, and the consequence of the assignment. Do not make the dispatcher traverse a marketing-style sequence or a new shell to finish a routine task.

Verify. Walk through the task with representative long and missing records. Check that the essential action remains understandable when optional content is slow or unavailable. Record the result rather than claiming a performance improvement without measurement.

Do not import. No additional shell, feature hero, framework migration, or new dependency. Keep the approved surface role.

Open the product skill · Evidence: Design Engineer Principles; How we think about design; Component Architecture for React Server Components

Preserve the place as the collection grows

Large customer, property, and inspiration collections become expensive to navigate when every inspection loses the current place.

Treat filters, selected records, scroll, and Back behavior as part of the task. Loading a child or optional panel should not reset the whole collection. A saved reference includes enough context to be useful when revisited.

Thorbis example. Open a property from a filtered customer list and return to the same filter and position. On the inspiration wall, reserve the media footprint and retain source credit even when the preview is unavailable.

Verify. Try refresh, Back and Forward, a direct record link, an empty result, a failed preview, and a long collection. Check that late data does not overwrite a draft or move the action under the pointer.

Do not import. No unbounded eager media, private information in shareable URLs, or a local-only storage promise borrowed from Copper.

Open the ecosystem skill · Evidence: 7 Principles of Rich Web Applications; Component Architecture for React Server Components; Philosophy; Copper

Loading search…

Loading search. You can close this message and keep reading.