Skip to main content
Tools

Tools we’ve built.

Open-source Claude Code skills shipped from this portfolio. The shiploop discipline that turns Claude into a shipping protocol, and the design skill that keeps generated UI inside your own system. MIT, no telemetry, drop into any .claude/ directory.

claude-shiploop

The eight-skill discipline — a lifecycle for shipping AI-collaborated software without churn.

GitHub ↗

Seven slash commands plus a domain-expert subagent that turn Claude Code from a code generator into a shipping protocol. Classifies every request, gates every build behind a spec, and leaves receipts (DEVLOG, memory, retro) in your repo — not in a chat window.

Install
curl -sSL https://raw.githubusercontent.com/ruslan4427/claude-shiploop/main/install.sh | bash

Writes to ~/.claude/{skills,agents,templates}. Prefer clone-and-inspect if auditing first.

Effectiveness

Why it exists
Claude without discipline codes immediately on every request. Discipline — classify, spec, plan, task, build, retro — is what separates vibes from engineering.
What it systematizes
Every incoming request runs through the 10-lane classifier before touching code. Every L-class feature produces spec+plan+tasks on disk. Every build iteration appends to DEVLOG. The protocol is the artifact, not another chat log.
Measured impact
Discipline
100%

On tier-2 (ShipLoop) runs: spec + plan + tasks + DEVLOG present.

Method · 4-point binary: spec/plan/tasks exist in specs/<N>/ and DEVLOG.md is present.

Baseline · Tier-0 projects (pre-ShipLoop) score 0% — none of the four artifacts exist.

Rework ratio
0.39

This portfolio (hrekov-dev) under ShipLoop vs 4.5 on a tier-0 baseline.

Method · fix(…) commits ÷ feat(…) commits (conventional commits only, lifetime).

Baseline · SMM Factory tier-0 = 4.5 — ten times more rework per shipped feature.

Expert hit rate
67%

First validated capsule caught 2 of 3 blind-bugs with plan-changing findings.

Method · (expert consulted · plan changed = yes) ÷ (total consultations). Hand-counted today; DEVLOG convention not yet wired for auto-rollup.

Baseline · Random guess on a 3-bug test ≈ 33% — the capsule doubled that.

→ Inspect the parser — the ground-truth source for every number above.

ui-ux-pro

A UI/UX skill that extracts your project's tokens first — then brings in current design thinking without breaking your system.

GitHub ↗

Two jobs. Keep a project's own design system consistent (extract tokens from CLAUDE.md / @theme / :root / tailwind.config, work inside them, never invent a new palette for a project that already has one). And bring in current design thinking — curated palettes, typography pairings, trend references — behind an opt-in style selector (minimalist / brutalist / Lusion).

Install
git clone https://github.com/ruslan4427/ui-ux-pro.git ~/.claude/skills/ui-ux-pro

No network at runtime. Two Python helpers for contrast and token extraction.

Effectiveness

Why it exists
AI design output drifts toward generic SaaS unless something forces it to read the project's existing system before suggesting a palette or font.
What it systematizes
Four-section method: tokens first, contrast gate, visual verification in the browser, then current design thinking. Style overrides are opt-in — defaults aim for a broadly safe editorial result.
Measured impact
—

No retroactive measurement — the skill shipped alongside this portfolio redesign but there's no clean pre/post session to compare. Will backfill when a measurement rig lands.

Inside ShipLoop.

Seven skills plus a domain-expert subagent. Each is a single-purpose SKILL.md that Claude Code loads on invocation. Not frameworks, not plugins — just the system prompt and the arguments.

Install
curl -sSL https://raw.githubusercontent.com/ruslan4427/claude-shiploop/main/install.sh | bash

Pipes through bash — prefer clone + inspect first if you want to audit the installer before running it.

Entry point

2

/shiploop-start

Single entry point. Classifies the current state (new project, new feature, continuing work, or status check) and routes to the right phase so you never guess which skill fires first.

Invocation
/shiploop-start new-feature description="<text>"
When to use

First command of any session. If you don't know whether the ask is a spec, a build, or a bug — run this.

/shiploop-router

Classifies any incoming request against the 10-lane table (Quick / Bug / Small / Large / Refactor / Testing / Discovery / Memory / Observer / Expert) and prints the matching protocol.

Invocation
/shiploop-router <user request text>
When to use

You caught yourself about to code before classifying. Pipe the message in and let the classifier call the lane out loud.

Lifecycle

1

/shiploop-brainstorm

Structured brainstorm at a defined lifecycle stage. Always produces: decisions, open questions, action items, and a memory entry — never a loose doc you have to re-read to find the call.

Invocation
/shiploop-brainstorm stage=discovery
When to use

Discovery before spec. Post-spec stress-test. Post-QA retro. End-of-release retrospective. The stage argument gates which prompts fire.

Quality + review

3

/shiploop-tester

Full QA pipeline: automated tests, Nielsen-10 UX heuristics, analytics sanity, prioritized fix tasks, and a QA report. Can scope to tests-only, ux-only, analytics-only, or a specific user story.

Invocation
/shiploop-tester tests-only
When to use

After a build iteration, before merge, or when a feature feels off and you want structured eyes instead of a vibe check.

/shiploop-observer

Audits the ShipLoop chain itself. Identifies dead agents, overlapping skills, missing lanes. Proposes consolidation or new skills — not another feature, but a workflow upgrade.

Invocation
/shiploop-observer full
When to use

The workflow feels slow or over-engineered. Run after every major phase to keep the ecosystem tight.

/shiploop-expert

Consults a domain-expert subagent at a defined review phase. Loads a cached capsule from .claude/expertise/<domain>.md or researches a fresh one (hybrid-approved) and returns issue → why → fix findings — never generic advice.

Invocation
/shiploop-expert phase=pre-ship
When to use

Pre-ship second opinion before merge. In-progress sanity check after plan lands. Blind-bug when symptom is clear but root cause is not. 3/3 blind-bug catch rate on first validated capsule.

Meta

1

/shiploop-skill-builder

Meta-skill for authoring and auditing other skills. Produces a valid SKILL.md for a new agent, or audits existing skills for structural compliance (frontmatter, section order, invocation hints).

Invocation
/shiploop-skill-builder create name=<skill> purpose="<one-liner>"
When to use

You spot a recurring workflow across sessions and want to codify it. Or audit before publishing a skill to a shared repo.