agentsclimarketplace

Pmo design

Skill witchwasin/Axiom-PMO/skills/pmo-design

Use for product flows, UX and wireframes, creative direction, visual direction, art direction, brand conformance, distinctive or non-generic frontend direction, design systems, visual sheets, and design-ready acceptance criteria tied to scoped requirements.From its SKILL.md

Install
npx -y skills add witchwasin/Axiom-PMO --skill pmo-design

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 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.

SKILL.md

9.6 KB, as published. Nobody here has run it

pmo-design

Purpose

Turn scoped requirements into flows, a source-backed visual direction, and a buildable design contract without inventing scope or approval.

Intents

IntentUse whenRead set
flow (default)Map requirements to flow, wireframe, or acceptance wordingPROJECT.md, relevant DESIGN/**
visual_directionEstablish creative/art direction, avoid generic defaults, or choose how the product should feelsee Required Inputs
design_systemTurn a selected direction or existing brand into a developer contract and visual sheetsee Required Inputs
system_designDefine architecture, data/API boundaries, and early Test Strategy in DESIGN/BUILD-SPEC.mdPROJECT.md, DESIGN/BUILD-SPEC.md
design_providerPrepare the governed Claude Design handoff, run preflight, reconcile candidate output, and record the reviewPROJECT.md, DESIGN/BUILD-SPEC.md, DESIGN/CLAUDE-DESIGN/**, EXTERNALIZATION.json, decision-log.md

Read docs/concepts/visual-direction.md for creative-direction rules and docs/concepts/design-system.md for the artifact contract. Do not restate them here.

Routing and Fallback

  1. Route requests for creative, distinctive, premium, unconventional, art direction, visual direction, or "not generic" work to visual_direction, even when the user also says "design system".
  2. For a new design system, consume DESIGN/VISUAL-DIRECTION.md when its direction_status is selected or conformance.
  3. If no usable direction exists, use visual_direction first. An existing design system may be maintained without retroactively creating this artifact unless the direction is being reopened.
  4. Never reopen a selected direction inside design_system; record a proposed change as an open question and return to visual_direction.

Required Inputs

Always read PROJECT.md, relevant requirements and design files, and the configured context/artifact contracts.

  • visual_direction: read DESIGN/VISUAL-DIRECTION.md if present, DESIGN/WIREFRAME.md, DESIGN/BRAND/**, decision-log.md, and only the configured source files needed for the brief.
  • design_system: additionally read DESIGN/VISUAL-DIRECTION.md, DESIGN/WIREFRAME.md, DESIGN/BRAND/**, and decision-log.md if present. Read only the scoped set.

Avoid delivery and release documents unless checking impact. The design system and visual direction remain optional in every mode; no gate requires either artifact.

Evidence and Authority

Use the repository evidence ladder exactly: verified = direct source plus human approval; supported = direct source without final approval; inferred = reasoned from partial source; missing = not found; conflict = sources disagree. Never use evidence status to describe tool availability or whether a render ran.

Only a human may declare brand_starting_point: undecided | existing, select an art direction, or approve Design Ready. Use direction_status: pending | selected | conformance; selected requires explicit human choice and a decision reference, while conformance requires a human-confirmed existing brand.

Execution - flow

  1. Map requirement IDs to user and system flow steps.
  2. Create or update flow and wireframe artifacts only when the configured contract or confirmed impact requires them.
  3. Record assumptions, open questions, and usable delivery references.

Execution - system_design

  1. Complete System Design and the mode-aware Test Strategy in BUILD-SPEC before detailed UI work.
  2. Create flow/wireframe artifacts only for an active UI path and actual UI scope.
  3. Record a technical/scope mismatch as a candidate governed Change Request; do not silently widen the approved contract.

Execution - design_provider

When UI delivery: claude_design is declared (see docs/concepts/claude-design-workflow.md):

  1. Prepare DESIGN/CLAUDE-DESIGN/INPUT-MANIFEST.json from templates/DESIGN-PROVIDER-INPUT.json: digest-bound references to the minimum necessary canonical inputs only — never source/** by default — plus an approved EXTERNALIZATION.json entry, provider, purpose, and generated timestamp. Run the designProviderDigest tool (src/tools/digest-tools.ts) and record the current digests.
  2. Hand the folder to the Human-operated provider; candidate output lands in DESIGN/CLAUDE-DESIGN/OUTPUT/**.
  3. Run the deterministic preflight (paths, manifest, declared screens/states, digests, scope references) before any review is recorded.
  4. Perform semantic candidate reconciliation for business/API/data/permission mismatches before Human review; these are candidate findings, not decisions.
  5. Record the Human review outcome in DESIGN/CLAUDE-DESIGN/REVIEW.json with a decision reference. Only a Human may mark acceptance (DPROV-006); an AI may propose revision.
  6. Route technical/scope findings through CHANGE-REQUESTS.json (DPROV-007) before the accepted design baseline is finalized; never silently widen the approved contract.
  7. After revision, recompute all digests; a changed output set invalidates the prior review. Accepted current output is what merges into Final Handoff.

Execution - visual_direction

  1. Read confirmed scope, decisions, wireframes, brand assets, and relevant sources before asking questions.
  2. Fill DESIGN/VISUAL-DIRECTION.md from the template. Keep source facts and design judgements at their truthful evidence statuses.
  3. Ask a human for brand_starting_point if it is not already human-confirmed. Do not infer or default it.
  4. For existing, assess conformance and use direction_status: conformance; do not perform a theatrical multi-direction exercise.
  5. For undecided, propose two or three directions that differ in geometry, typography, composition, density, colour role, imagery, surface/depth, or motion—not merely hex values. Set direction_status: pending, then stop for human selection.
  6. After an explicit human choice, set direction_status: selected, record the decision reference, and carry the selected direction into design_system. Never select on the human's behalf.
  7. If editing decision-log.md, first check whether it is semantic-review input; editing it can stale HANDOFF-REVIEW.json.

Execution - design_system

  1. List only screens already scoped in PROJECT.md and DESIGN/WIREFRAME.md.
  2. Consume the selected/conformance direction and explain every presentation choice by pointing to a brief field. If that sentence cannot be written, the choice is an unexamined default.
  3. Write brand assets when needed, then DESIGN/DESIGN-SYSTEM.md and the self-contained HTML sheet from their templates. Keep the first token block and required semantic anchors intact; rewrite the presentation layer to fit the direction.
  4. Keep Markdown and HTML token values identical. Fill the Sample Data Register and measure every contrast ratio stated.
  5. Render and show the sheet when tools permit. Report tool_availability and render_status separately; never claim a render or review that did not happen.
  6. At Handoff, when VISUAL-DIRECTION.md, DESIGN-SYSTEM.md, and DESIGN-SYSTEM.html all exist, prepare honest desktop and mobile captures when rendering tools permit. A named human must review them and record DESIGN/VISUAL-REVIEW.json; an AI may draft notes but may not fabricate the reviewer, decision reference, or aesthetic outcome.
  7. Report assumptions, open questions, scoped screens drawn, undrawn states, and meaningful decisions. Editing a semantic-review or Visual Proof input can stale its candidate evidence.

Output Contract

Return changed design files, requirement coverage, evidence status, assumptions, unresolved questions, and any human decision still required. For visual work, also report selected/conformance direction, screens and states drawn or missing, tool_availability, render_status, and the rendered path when one exists. When Visual Proof applies at Handoff, report whether its capture and named-human review evidence are missing or stale; do not turn that report into approval.

Prohibited Actions

Do not expand scope, remove source references, invent requirements or decision IDs, infer brand_starting_point, select a direction, mark approval, draw an unscoped screen, present an inference as confirmed, use real customer or personal data, claim an unavailable tool ran, or publish externally without explicit current-session instruction.

Completion Criteria

  • flow: every design-affecting item has a design reference or an allowed not_required sentinel.
  • visual_direction: brief evidence is truthful; status is pending, selected, or conformance consistently; any selection has human evidence and a decision reference.
  • design_system: the direction is traceable, DESIGN-001 passes, no HTML placeholder remains, sample data is registered, every screen traces to a wireframe, and render/tool claims are honest. At Handoff, the conditional Visual Proof contract is complete only after a named human has recorded both committed captures and the review manifest.

Validation

node cli/axiom.mjs validate --project <project> --mode <mode> --gate Design

The artifact pair is candidate evidence only. Design Ready remains human-owned.

Visual Proof stays in the repository workflow: active instructions are under .claude/skills/, and the generated skills/ package is refreshed by its build command. Do not create or maintain a separate .agents/skills mirror for this capability.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.