agentleFS
igovetigovet/cortex-orch-acp/skills/design-workflow/SKILL.md

design-workflow

Skill1 starsChanged 59 days agoMITView on GitHub

Coordinate adaptive UI/UX intake, art direction, implementation and evidence-backed visual review using existing Cortex roles, without a separate lifecycle or design store.

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 10 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 10

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use for new UI/UX design, a visual redesign, a design system or significant visual improvement.

    line 10
  2. Very specific:

    Ask only unanswered material design questions: product goal, audience, platforms, screens/content/states, visual direction, references and …

    line 26
  3. Distinctive:

    When documents are missing or insufficient, plan a technical-writer assignment to record agreed answers into PRODUCT, DESIGN and SCREENS …

    line 33
  4. Distinctive:

    Read linked package references through Cortex read_skill using design-intake, PRODUCT, DESIGN, SCREENS or visual-review, not target-project …

    line 39
  5. Distinctive:

    Prefer originality and usability over generic card grids or decorative trends.

    line 52
  6. Distinctive:

    Read-only roles consume writer-captured images.

    line 70
  7. Distinctive:

    Use only the session-listed skill for the relevant stage, read on demand.

    line 79
  8. Distinctive:

    Do not load every skill for every worker.

    line 91
  9. Distinctive:

    Use Playwright MCP only if its tools are actually advertised and usable in the current session.

    line 97
  10. Distinctive:

    If unavailable, use the project's supported screenshot tooling or user-provided captures.

    line 101
  11. Distinctive:

    No model self-score, passing build or unopened screenshot proves visual quality.

    line 116

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Design workflow

Activation

Use for new UI/UX design, a visual redesign, a design system or significant visual improvement. Do not activate for backend work, nonvisual bug fixes, routine frontend coding or purely functional UI changes without design decisions. A routine change normally needs one frontend-dev, not a mandatory design team.

This skill informs the plan; the runtime still owns scheduling, permissions, independent code review, verification and completion. Do not create agents, parallel orchestration or a new store.

Intake and durable context

Build on requirements-intake when material inputs are missing. First read the target project's instructions and docs/PRODUCT.md, docs/DESIGN.md and docs/SCREENS.md when present, plus established tokens and components. Respect equivalent existing documents; do not duplicate or replace an approved design system blindly.

Ask only unanswered material design questions: product goal, audience, platforms, screens/content/states, visual direction, references and what they demonstrate, avoid list, constraints and observable success criteria. Use grouped structured questions, meaningful options and free text, at most two rounds. Missing files alone do not justify questions if the request and code already answer them. See the design intake bank.

When documents are missing or insufficient, plan a technical-writer assignment to record agreed answers into PRODUCT, DESIGN and SCREENS before design work, preserving existing facts and clearly identifying unresolved choices. Never invent functionality. Templates: PRODUCT, DESIGN, SCREENS.

Read linked package references through Cortex read_skill using design-intake, PRODUCT, DESIGN, SCREENS or visual-review, not target-project file paths.

Adaptive phases and ownership

Scale the phases to design risk, not to a fixed five-worker ritual. Reuse approved direction; omit redundant research or documentation work when already sufficient. For substantial visual change, make evidence and independent visual/accessibility review explicit in objectives and dependencies.

  1. Art direction — ux-designer, read-only. Research available references; return audience-specific principles, flows, type/color/spacing tokens, component states, responsive rules and rationale in the report. Do not write DESIGN.md. Prefer originality and usability over generic card grids or decorative trends. Apply the hard bans in the DESIGN reference's "Banned AI-tell patterns" section.
  2. Persist — technical-writer or assigned writing role. Consume the design report and record the chosen direction in docs/DESIGN.md, keeping PRODUCT and SCREENS aligned. Dependent implementation consumes that exact approved spec; do not make developers silently invent a different direction.
  3. Implement and capture — frontend-dev or mobile-dev. One active writer per screen/shared UI files. Follow the approved direction and platform conventions. Capture the actual render at mobile, tablet and desktop sizes relevant to the product; native mobile uses supported device classes, with desktop explicitly not applicable where justified. Record exact dimensions and platform. Save screenshots in a project-local git-ignored evidence path outside private storage (for example .design-evidence/ after verifying its ignore rule). A qa-engineer writing assignment can capture evidence if needed, not restyle UI.
  4. Independent visual review — ux-designer plus accessibility-auditor. Use a fresh review assignment, not the implementation author's self-review. Compare opened screenshots with DESIGN and SCREENS; report concrete defects, user impact, severity, state, viewport and remediation. Accessibility review distinguishes screenshots from exercised keyboard/focus/assistive-technology evidence. Read-only roles consume writer-captured images; they do not create files. This visual review supplements, never replaces runtime mandatory code review.
  5. Refine — the same implementation owner. Return actionable findings to that owner, then recapture/review affected states and viewports. Plan at most two refinement rounds; if material defects remain, report them and request a scope decision or replan rather than endlessly spawning new owners.

Stage-to-skill map

Use only the session-listed skill for the relevant stage, read on demand.

StageSkillOwners
Reference researchdesign-reference-analysisux-designer
Design systemdesign-ui-ux-pro-maxux-designer, mobile-dev
Implementationdesign-frontend-designfrontend-dev, mobile-dev
Critique / refinementdesign-impeccableux-designer, accessibility-auditor (review); frontend-dev (edits)
Capture / rendered reviewdesign-visual-qafrontend-dev, mobile-dev, qa-engineer (capture); ux-designer, accessibility-auditor (review)

Project-local .agents/skills extras take precedence when session-listed and applicable, subject to project instructions, authorization and the agreed brief. Do not load every skill for every worker. Skill opinions and reference tables never override the product's approved direction. Derive principles from references, not copied markup, brand assets or one-to-one layouts.

Browser and evidence boundary

Use Playwright MCP only if its tools are actually advertised and usable in the current session. Configuration is not availability. Check current tool discovery; do not assume a browser based on a previous session. Read interactive-resources before using a browser/server/device resource and preserve its single owner. If unavailable, use the project's supported screenshot tooling or user-provided captures. If neither exists, name the missing capability and mark rendered claims unverified; do not infer appearance from code or install tools without authority.

The bundled skill does not install browsers, repair host registries, or promise that a particular MCP server exists.

Visual quality gate

Apply the visual review checklist. Evidence must bind to the implemented revision, screen, state and viewport. Require opened mobile/tablet/desktop captures (or justified platform exclusions), typography and hierarchy, composition/spacing, responsive reflow, required content/states, usable interactions and accessibility checks against DESIGN/SCREENS.

No model self-score, passing build or unopened screenshot proves visual quality. Retain concrete findings and same-state rechecks. Missing evidence, hard failures and unresolved material findings stay explicit blockers, not diagnostic passes. The coordinator synthesizes these facts; only the runtime can accept completion.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.