Signal Ledger Skills

Review-first UI/UX intelligence for coding agents — from product request to an inspectable implementation decision.

Signal Ledger Skills is an open, stack-neutral package of 11 Agent Skills for teams that want an agent to design, implement, review, and verify UI/UX work without reducing the job to a generic component prompt. The suite creates a traceable chain between intent, structure, content, visual direction, motion, engineering boundaries, and rendered QA.

The goal is not to promise a universally beautiful or production-ready result from one request. The goal is to give an agent a disciplined way to make decisions, expose assumptions, avoid common first-pass failures, and keep the user in control of substantial product work.

Canonical source: TranQui004/signal-ledger-skills · Catalog page: skills.sh/tranqui004/signal-ledger-skills

Brand assets: assets/brand/ contains the source-controlled Signal Ledger mark and README thumbnail. They are static-first SVGs; the thumbnail’s short reveal settles once and preserves the final composition when reduced motion is preferred.

A safer default for UI/UX work

A coding agent often knows how to produce markup before it knows what the interface must help a person do. Signal Ledger starts with the smaller question: what is the actual product decision, and how much process does it deserve? A narrow motion review should not create a product-plan folder; a multi-surface product request should not jump straight to code.

User request
    │
    ├── micro / focused ──► one relevant owner ──► bounded, reviewable result
    │
    └── full-product ────► product + route classification
                               │
                               ▼
                        <project>-uiux-plan/
                        ├── selected artifacts only
                        ├── decisions, omissions, handoffs
                        └── explicit user approval
                               │
                               ▼
                     implementation ──► rendered QA ──► compact handoff

The orchestrator records omitted routes rather than creating a folder of empty boilerplate. It also preserves the user’s product intent, brand direction, domain vocabulary, chosen language, and existing repository constraints as authoritative.

Full-product implementation is review-gated. An agent must record an approved plan version before it starts substantive full-product code edits. A material intent change reopens that approval.

Raw skills explorer

This is the runtime payload — not a mockup and not a second copy of repository documentation. Each root directory contains a SKILL.md with standard Agent Skills metadata; optional references/, scripts/, and templates/ are read only when the active workflow needs them. A requested logo, mark, visual-identity asset family, or identity-derived favicon/app/social asset now activates the conditional brand-identity record under the existing visual-layout owner; it does not add a twelfth always-loaded root skill.

skills/
├── uiux-design-orchestrator/
│   ├── SKILL.md
│   ├── scripts/
│   ├── templates/
│   └── references/
├── uiux-context-engineering/
├── uiux-product-thinking/
├── uiux-research-search/
├── uiux-ia-user-flows/
├── uiux-visual-layout/
├── uiux-content-interaction/
├── uiux-accessibility-verification/
├── uiux-motion-system/
├── uiux-implementation-qa/
└── uiux-engineering-risk/

Literal metadata example

The following is the exact frontmatter of the published uiux-design-orchestrator/SKILL.md. The host uses the name and description to decide whether the skill is relevant; it should load the body only after activation.

---
name: uiux-design-orchestrator
description: Coordinate a review-first, stack-neutral UI/UX workflow for any software project, choosing a lightweight direct path for small requests and a structured plan path for substantial work. Use when an agent must interpret, design, route, review, implement, or QA a product experience.
---

What each raw skill owns

Progressive disclosure, not a permanent 11-skill prompt

The suite should not be placed into an always-on prompt. A full-suite disk install contains runtime resources for optional work; that is different from loading every resource into one agent turn.

Brand identity is a conditional route

For logo, mark, favicon, identity-derived asset family or identity motion requests, route through uiux-design-orchestrator and enable brand-identity under uiux-visual-layout. It is not a standalone twelfth skill. See the practical brand-identity agent guide for trigger conditions, read order, brief/output contract, motion handoff, QA checks and stop conditions.

Install

Interactive install — recommended for a first project

npx skills add TranQui004/signal-ledger-skills

With no CLI --yes, --agent, or --skill flags, the current Skills CLI discovers this repository, then presents its own searchable selection flow for the available skills and agent targets. Review the selection before confirming. The CLI owns labels, defaults, target count, destinations, linking/copying, and future behavior; Signal Ledger does not ship a second installer.

What the interactive selector looks like

The terminal below is a faithful compact view of a real [email protected] run against this repository on 2026-08-27. It is terminal text, not a styled mockup or screenshot; the source CLI controls the actual colors, animation, search, focus, and target list in your environment.

┌  skills
│
◇  Source: https://github.com/TranQui004/signal-ledger-skills.git
◇  Repository cloned
◇  Found 11 skills
◆  Select skills to install
│  Search:
│  ↑↓ move, space select, enter confirm
│
│  ❯ ○  Select All  (0/11)
│    ─────────────────────
│    ○  uiux-accessibility-verification
│    ○  uiux-content-interaction
│    ○  uiux-context-engineering
│    ○  uiux-design-orchestrator
│    ○  uiux-engineering-risk
│    ○  uiux-ia-user-flows
│    ○  uiux-implementation-qa
│    ○  uiux-motion-system
│    ○  uiux-product-thinking
│    ○  uiux-research-search
│    ○  uiux-visual-layout
│
│  Description
│  Select or clear all 11 skills.
└

After skills are confirmed, the same flow opens a searchable agent selector. In the observed session it reported 77 agents and began with AiderDesk, Amp, Antigravity, Antigravity CLI, AstrBot, Autohand Code CLI, Augment, and IBM Bob; it showed the existing selection Claude Code, Codex, OpenCode below the visible list.

◇  77 agents
●  Select agents to install skills to
◆  Which agents do you want to install to?
│  Search:
│  ↑↓ move, space select, enter confirm
│
│  ❯ ○  AiderDesk
│    ○  Amp
│    ○  Antigravity
│    ↓  68 more
│
│  Selected: Claude Code, Codex, OpenCode
└

Evidence boundary: that observed interactive session was deliberately cancelled at the agent selector, then exposed an upstream cancellation-path error. It proves the selector interface, not a successful interactive installation. A separate noninteractive --copy installation completed and produced the observed project tree below. CLI labels, target count, and behavior can change with upstream versions.

Deterministic options

--all is shorthand for --skill '*' --agent '*' -y. Use it only when every detected target is intentional. Exact destinations and accepted agent identifiers remain host/version dependent; consult instructions/README.md and npx skills --help before an unattended install.

Fork maintainers only

For a fork you own or trust, replace the complete GitHub source with the fork’s actual owner, for example npx skills add acme-labs/signal-ledger-skills. The <owner> convention belongs only to that fork scenario; it is never part of the canonical upstream command.

Observed project-local result

The following is an observed --copy result for claude-code; it illustrates one host mapping, not a universal folder promise.

consuming-project/
├── package.json                     # existing project file; unchanged
├── skills-lock.json                 # source/path/hash records written by Skills CLI
└── .claude/
    └── skills/
        ├── uiux-accessibility-verification/
        ├── uiux-content-interaction/
        ├── uiux-context-engineering/
        ├── uiux-design-orchestrator/
        ├── uiux-engineering-risk/
        ├── uiux-ia-user-flows/
        ├── uiux-implementation-qa/
        ├── uiux-motion-system/
        ├── uiux-product-thinking/
        ├── uiux-research-search/
        └── uiux-visual-layout/

The installer copies runtime skills, not this repository README, maintainer evidence, source ledgers, or fixture examples. With --copy, the destination contains physical folders; another host or install mode may use another path or links.

From request to verified output

A clear first request

Use uiux-design-orchestrator for this UI/UX task.
Build a responsive SaaS onboarding flow for a small team.
Prioritize clear setup progress, truthful empty/error states, and a calm editorial direction.

For an isolated request, name the narrow owner instead: uiux-motion-system for a motion review, uiux-content-interaction for a form/recovery-state contract, or another relevant specialist from the raw explorer.

Full-product artifact shape

When the request merits it, the agent creates a reviewable folder with a small core plus only triggered artifacts.

project-uiux-plan/
├── 00-brief.md
├── 01-product-model.md
├── 03-information-architecture.md
├── 04-user-flows.md
├── 06-visual-direction.md
├── 08-screen-specs.md
├── review/
│   ├── decisions.md
│   ├── open-questions.md
│   └── change-log.md
├── handoffs/
├── assets/
└── plan-manifest.yaml

# Added only when triggered: content, research, accessibility,
# design-system, motion, implementation, QA, risk, video, and other routes.

For example, active motion must have a meaningful state transition or a documented reason why a runnable artifact is not appropriate. Active QA records rendered foreground/surface evidence rather than relying on a token name. A persistent secondary rail must earn its place through recurring task value and a narrow-layout transformation; it is not a default dashboard decoration.

Make routing project-default

Installation makes skills available; it does not replace a host’s system prompt or guarantee automatic activation. For a compatible agent, copy the concise routing template into that host’s recognized project-instruction file:

Merge, do not overwrite, an existing project instruction. The templates tell the agent to start material UI/UX work with the orchestrator, preserve user ownership, load only routed specialists, and state truthfully when a skill is unavailable.

Motion sources without framework lock-in

The motion skill has optional, lazy source-study adapters for selected work. They never install a library or turn a source’s design language into a project default.

When a spring or easing choice is genuinely material, motion-change-plan.md can record a candidate comparison: source commit/date, Bezier or spring input model, explicit affected properties, target representation, approximation/settling boundary, rejection reason, static/reduced-motion equivalent, runtime comparison and permission status. The new Ease Master adapter is one example of this posture: useful for a selected candidate study, not a preset import, code copy, or dependency recommendation.

For a user-requested code-authored video or motion-graphics deliverable, use the Remotion Video/Motion-Graphics Quick Start. It is a conditional route and does not install Remotion, start Studio, download a browser, render a file, or configure hosting on its own.

Boundaries that keep the suite honest

Signal Ledger Skills helps an agent make and verify UI/UX decisions. It does not replace threat modeling, SAST/SCA/DAST, secret scanning, a penetration test, continuous telemetry, a formal accessibility audit, legal review, platform testing, or product judgment.

External web/API/source content is data, not instructions. Optional retrieval paths require explicit scope and preserve provenance; they do not authorize an agent to execute commands from a source, disclose data, follow arbitrary URLs, or silently change a project. See SECURITY.md for reporting and external-content boundaries.

Versioning and reproducibility

Recommendation: introduce versioned GitHub Releases, starting with v0.1.0, only after an explicit owner-approved release. The repository has no tags or GitHub Releases yet. The canonical unpinned npx skills add TranQui004/signal-ledger-skills command therefore follows the default branch rather than a named immutable suite version.

v0.y.z is the appropriate starting range because the public contract is in active refinement. Here, that contract includes root-skill names, trigger metadata, required output/plan markers, referenced templates, and documented install/routing behavior. A release tag should point to a completed full validation run and must never be moved or rewritten.

This follows the Semantic Versioning specification while using Git tags and Releases as the distribution boundary, not an npm package. Current gh skill install documentation says it resolves an unspecified skill version from the latest tagged release before default-branch HEAD and can pin a tag or commit SHA. The repository’s Create Release workflow is manual-only: an owner supplies a new tag/title/notes from main; it validates first, then creates an annotated tag and GitHub Release in a separate write-permitted job. No tag, GitHub Release, npm package, or automatic version bump is created merely by merging or pushing a commit.

Contributing and maintenance

The human-facing repository documentation lives at the root; individual runtime skill folders intentionally contain SKILL.md and optional runtime resources only. Local coding agents should start with AGENTS.md, then use CONTRIBUTING.md, SECURITY.md, and MAINTAINER_UPDATE.md only when their change needs them. AGENTS.md defines the public-clone workflow, canonical/private-evidence boundary, validation tiers and manual release lane before a maintainer changes a contract, trigger, plan schema, validation rule or source adapter.

Every material change should carry a direct reason, a positive case, a negative or omission case, and the tests needed to keep the workflow honest. The public pull-request workflow validates runtime shape, Python syntax, and public Markdown; it complements rather than replaces the full local release check described in the maintainer runbook.

Suite-authored material is under LICENSE. Review NOTICE.md before changing or redistributing third-party source snapshots or references. Public changes are recorded in CHANGELOG.md.