A library for repeatable design-agent work.

A workflow that fans out one agent per component or screen to find token drift, spacing inconsistency, duplicated patterns, and undocumented variants, then aggregates everything into a prioritized drift report.

A dynamic workflow that captures every key page, state, and viewport, fans out one compare agent per page, merges the findings, and drives fix passes until the whole product matches the approved design again.

A single-agent build loop that turns a reference screenshot or Figma export into working code by extracting a spec first, building against it, and refusing to call the screen done until a parity check against the reference passes.

A workflow for migrating hard-coded values and legacy tokens to a new token set across hundreds of files, with a script that loops per-file agents, a reviewer agent on every diff, and a visual QA recapture at the end.

A codebase-wide accessibility audit that combines static checks with agent judgment, ranks findings by severity, and applies fixes in passes that are re-verified before anything is called done.

A parallel-agent workflow that drafts three to five genuinely distinct design directions from one brief, forces an adversarial critique stage where agents score each other's work against explicit criteria, and hands a human a comparable set of options instead of one early draft.

A saved Claude Code workflow command that runs design QA on every branch that touches UI: build the preview, capture only the changed screens, diff against main, check tokens and accessibility on the changed components, and produce a findings comment ready to paste into the pull request.

A subagent fan-out that turns approved designs into developer-ready specs: layout redlines, token mapping, interaction states, accessibility annotations, edge cases, and acceptance criteria, assembled into one handoff packet and reviewed from a developer's perspective before it ships.

A dynamic workflow that generates and refreshes design system documentation from the source of truth — component code, tokens, and real usages across the product — with one agent per component and a consistency agent enforcing a shared template and voice.

A repeatable intake pipeline for new component and pattern proposals: an overlap agent checks whether the need is already covered, an impact agent estimates adoption, a spec agent drafts the API and accessibility requirements, and a review packet goes to the design system council to decide.

A half-day loop that turns a brief and a design harness — tokens, components, sample data — into a clickable coded prototype, fans out variants only where the design question genuinely differs, and packages everything behind an index page for testing sessions.

A recurring workflow that assembles the design ops evidence teams currently gather by hand — design system adoption, QA debt trends, review throughput, and contribution backlog — into one report where every number links to how it was computed.

A pre-localization review that finds the internationalization risks hiding in UI and code — hard-coded strings, fragile layouts, formatting assumptions, and RTL gaps — before translators and in-market reviewers ever see the product.

A workflow for running a structured heuristic evaluation across every key flow of a product at once, with one agent per flow working from screenshots and the live build, and a merge agent that dedupes and ranks the candidate violations for human severity review.

A two-stage workflow that prepares a usability study — research questions, test plan, tasks, screener, discussion guide, prototype check — and then synthesizes the sessions afterwards, with one coding agent per session, a synthesis merge, and a challenge pass that hunts for over-claims.

An evidence-based journey mapping workflow: per-source agents extract moments and signals from analytics, tickets, session notes, and interviews, a synthesis agent assembles stages with every claim linked to its evidence, and a challenge agent flags the stages where the evidence is too thin to trust.

A workflow for assembling a service blueprint — customer actions, frontstage, backstage, support processes, and evidence — from research repositories, support tickets, ops runbooks, and stakeholder notes the organization already has, so the validation workshop starts from a draft instead of a blank wall.

Two connected loops for information architecture work: an audit that crawls the navigation and content inventory with one agent per section, and a card sort and tree test analysis stage where agents compute agreement and label confusion while a human IA designer judges the tradeoffs.

A workflow for running structured cognitive walkthroughs across many tasks and personas at once: each agent steps through one task with one persona's knowledge level, answers the four classic walkthrough questions at every step with screenshot evidence, and a merge agent groups failure points by screen so the team sees where journeys break.

A staged workflow that prepares a design or discovery workshop from material that already exists: an evidence pack with sources, a stakeholder map, a timed agenda with facilitation notes, and ready-to-use templates — all reviewed and owned by the facilitator.

A workflow for using Claude Code's bundled /deep-research dynamic workflow to answer design questions with cross-checked, cited evidence: competitive teardowns, pattern research, accessibility standards, and platform conventions.

A workflow for turning a folder of interview transcripts, support tickets, or survey answers into coded themes, evidence-linked insights, and an opportunity map, with one agent per transcript and a challenge agent hunting for over-claims.

A workflow for tearing down competitor products by walking them — screenshots, flows, pricing, positioning, and onboarding collected into per-competitor evidence folders — then extracting patterns into a comparison matrix, a pattern library, and a list of opportunity gaps.

A workflow for sweeping product copy across buttons, empty states, errors, onboarding, and emails against a content style guide, with one agent per surface and findings merged into a terminology map and a prioritized fix list.

A two-stage workflow where a critique loop stress-tests the survey before anyone answers it, and a fan-out of analysis agents computes distributions, cross-tabs, and open-text themes from the export — with every number produced by code the agents run, never estimated by the model.

A staged workflow for designing A/B and multivariate experiments properly — hypothesis from evidence, sample size from a script, ship criteria agreed before launch — and producing honest readouts where a skeptic agent attacks the analysis before the team sees it.

A workflow that connects funnel numbers to the screens behind them: analytics agents compute step conversion and segment differences with scripts, walkthrough agents inspect the actual UI at the biggest drop-off points, and the output is a ranked list of evidence-linked hypotheses — never a causal claim from correlational data.

A parallel-agent workflow that turns one client brief into three to five campaign territories developed by separate concept agents, runs every territory through adversarial critique against the brief, brand guidelines, audience insight, and legal constraints, and hands the creative director a scored shortlist instead of one early idea.

A dynamic workflow that turns one approved master concept into a full matrix of channel assets — web hero, email module, paid social sizes, display banners, in-app banner — built by per-asset agents from a channel spec matrix, with every asset passing a brand-compliance gate before a human review board sees it.

A fan-out workflow that audits a landing page section by section — hero, social proof, pricing, form, footer — from screenshots and the DOM, checks message clarity, hierarchy, trust, friction, accessibility, and analytics gaps, and produces a prioritized backlog of test hypotheses rather than promised conversion wins.

A staged workflow that turns an approved identity direction into a usable system: contrast-checked design tokens, a living guidelines document, proof-of-concept applications built as code, and a consistency report — with the creative director reviewing every stage gate.

A research workflow that produces evidence-based content briefs for writers: per-query agents study what currently ranks and why, an intent agent classifies what searchers want, coverage and sources agents map the gaps and the references — and the writer and SEO lead make the calls.