Skip to content

ans-4175/interaction-design-skills / skills/interaction-design

interaction-design

Verified

Guide, audit, or decide human-centered interaction design — digital products (apps, web, dashboards, forms), information design, and services or organisational processes (requests, approvals, onboarding) — using named laws and heuristics (Fitts, Hick, Miller, Jakob, Tesler, Doherty, Peak-End, Norman, Nielsen, Gestalt, service blueprint). Use when designing a screen, flow or process, reviewing one for usability problems, or choosing between design options.

Interaction Design

Three modes: GUIDE (design something new), AUDIT (judge something that exists) and DECIDE (choose between options). Pick the mode from the request. If the request is ambiguous, say which mode you are using in one line and go on.

Ground every claim in a named principle and in the actual artifact (screen, flow, code, screenshot, form, SOP, process description). A law alone is not a finding: say where it applies and what changes.

References (read only what the task needs)

FileRead when
references/laws.mdAny mode. The core laws: what each says, when it applies, common misuse
references/heuristics.mdAUDIT. Nielsen's 10, Norman's principles, severity scale
references/patterns.mdThe work touches errors, feedback, loading, forms, navigation, search, onboarding, gestures
references/states.mdA component or flow has async work, branching, or many states
references/information-design.mdLayout, grouping, hierarchy, density, tables, charts, dashboards
references/service-process.mdServices and human processes: several people/units, approvals, documents, offline channels (service blueprint, handoffs, SLAs)

Choosing which principles matter

Start from the user's situation, not from the list of laws:

SituationLook first at
Too many options or a crowded screenHick, Miller (chunking), progressive disclosure, density
Hard to hit, mis-taps, mobile reachFitts
"Users don't find X"Jakob (convention), scent/wayfinding, serial position, von Restorff
User does work the system could doTesler, Postel, recognition over recall
Feels slowDoherty, loading ladder, optimistic UI
Errors, data loss, frustrationError hierarchy, Norman constraints, undo over confirm
Unclear what happened after an actionFeedback, visibility of system status, gulf of evaluation
Multi-step / long tasksGoal-gradient, Zeigarnik (with autosave), Peak-End
Grouping, alignment, "looks messy"Gestalt, visual hierarchy, aesthetic-usability
A process run by people/units (requests, approvals, intake)service-process.md: blueprint, handoffs, state ownership + Tesler, visibility of status

Apply only 2–5 principles to any one problem. Listing every law is noise.

GUIDE — design a screen, component, or flow

  1. Frame: who the user is, the primary task, the context (device, frequency, stakes), and what "done" looks like. Ask only for what blocks the design; otherwise state your assumptions.
  2. Structure: the content and actions on the screen, grouped and ordered by importance and task sequence (Miller, serial position, Gestalt).
  3. Interaction: for each key action give the trigger, the rules, the feedback and the result. Use the conventional pattern unless there is a stated reason not to (Jakob).
  4. States: empty, loading, partial, error, success, no-permission, offline where relevant (states.md). Each state needs a way out.
  5. Edges: errors and recovery, long or missing content, first use vs. repeat use, keyboard and screen-reader access.
  6. Output: a short spec. Cover structure, interactions, a state table, copy for key moments and open questions. Tag each design choice with the principle behind it, e.g. (Hick) or (Tesler). Include an ASCII wireframe of each key screen and the flow between them (see Wireframes).

AUDIT — review a screen, flow, or implementation

  1. Scope: which screens, flows or tasks are in scope. Walk the primary task as a first-time user, then as a frequent user.
    • Trace across roles. When a task spans people (trainer shows a QR → participant scans; requester submits → approver acts), walk both sides. The root cause often sits on the other role's screen, such as the wrong artifact being shown or shared.
  2. Inspect the real thing: the running app (drive it with a browser tool) and the code. Do not audit from memory. Live evidence finds what code misses (layout below the fold, stale messages, real copy); code finds states you can't trigger. Use both when you can.
  3. Check against heuristics.md, then the relevant laws and patterns. Pay extra attention to the unhappy paths: errors, empty states, slow network, expired session.
  4. Report findings ranked by severity:
[S3] <short title> — <where: screen/component/file:line>
Principle: <law/heuristic>
Problem: <what the user experiences, concretely>
Fix: <specific change>

Severity: S4 blocks the task or loses data · S3 major friction or a frequent error · S2 minor · S1 cosmetic. Put the strengths that must be kept at the end, in one or two lines. Do not pad the report: five real findings beat twenty generic ones.

For every S3/S4 finding, and any fix that changes layout or flow, add a before → after ASCII wireframe (see Wireframes). When findings become tickets, the wireframe goes into the ticket too, so the builder doesn't have to reconstruct the layout from prose.

Wireframes

Use plain ASCII inside a code block so it renders in chat, Markdown and issue trackers. Keep it at low fidelity: show structure, order, the primary action and the copy that matters; leave out styling.

  • Screen: a box at the target width (≈36 chars for mobile, ≈72 for desktop). Mark the fold with ┄┄ fold ┄┄ when position matters, show the primary action as [ Action ] and secondary ones as ( action ), and mark removed elements ~~like this~~ or with a ✕ note.
  • Flow: screen → screen steps on one line, with the trigger on the arrow when it isn't obvious (scan QR → login → [OTP] → name+confirm → ✓ done). Count the steps before and after.
  • One wireframe per decision. Don't draw screens that don't change.

DECIDE — choose between options

  1. Restate the options and the decision criteria: primary task, user type, frequency, stakes, platform, and constraints (time, engineering cost).
  2. Name the trade-off each option makes, in terms of the laws. Examples: fewer choices against more hidden functionality (Hick vs. discoverability); convention against differentiation (Jakob); simplicity for the user against effort for the product (Tesler).
  3. Score each option against the criteria in a compact table, giving reasons, not just numbers.
  4. Recommend one option. State what would change your mind, and the cheapest way to validate the choice (first-click test, 5-user test, A/B, analytics).

Rules

  • Conventions and thresholds (44pt targets, 400ms, 4±1 chunks) are design targets, not physics. Say so when a decision depends on one.
  • Accessibility is part of every mode. That means keyboard access, focus, contrast, never color alone, reduced motion, and labels.
  • Prefer undo over "Are you sure?" Keep confirmation for actions that are both destructive and irreversible.
  • When a recommendation conflicts with an existing design system or DESIGN.md, follow the system and flag the conflict.
  • Match the user's language in the output. Keep the law names in English.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall