8 skills · 25 entry points · one command

Agent skills for the work around the code.

A coding agent writes code well and does almost everything around it badly. It builds an interface with no idea who uses it, calls a task done without checking what was asked, and ships a page no answer engine can read. These 8 skills each take one of those gaps and give the agent a contract it has to follow — documentation, validators and small standard-library scripts. No services, no telemetry, no API keys.

install the whole family
$
npx sshlg-skills install
8skill packs, pinned and released together
25entry points an agent can be routed to
70+agents supported, Claude Code as plugins
0runtime dependencies, services or keys

The family

Each pack owns one question. They compose: the router decides the route, and every route names the phrase that declines it.

super-ux v0.48.0

what the interface must do and how it sounds

Scenario-driven UI development: a versioned design chain in docs/ux/ — the product vision, personas and jobs, user flows and the paid-acquisition funnel among them, a screens-and-states map with Figma frames…

7 skillsRead →

task-pipeline v1.77.0

how a change reaches the repository

Full-cycle delivery orchestrator: an intake grill turns the request into a complete brief, then ten gated stages carry it from docs to acceptance, refusing to advance until each gate passes.

3 skillsRead →

agent-sync v1.17.0

who is holding this file right now

Coordination plane for concurrent agents: leases with a TTL so two agents cannot claim the same work, race-free id reservation, a run journal and a generated board, over a pluggable knowledge cloud.

1 skillRead →

make-skill v0.23.1

how the skill or plugin itself is built

A skill that builds skills: create, retrofit, audit and publish agent skills and Claude Code plugins — Agent Skills standard conformance, marketplace layout, version sync, validator + CI, every distribution…

1 skillRead →

sheleg-design v1.49.1

how it looks and moves

The taste layer: cinematic scroll-driven landing pages (one scroll clock, motion that degrades to calm, WebGL particle formations), product-UI style packs with a ready token layer each, and the Figma border…

1 skill · a catalogue of its ownRead →

seo-aeo-audit v0.25.1

whether a machine will find it

Evidence-first website audit for search and answer engines: ten tracks from crawl access to AI citation mechanics, every finding backed by an observation and every recommendation tiered, ending in a…

1 skillRead →

sheleg-dev v0.10.0

integrations: money in, tracking, sign-in, speed

The integration layer a product reaches once it has users: Stripe subscription billing reconciled into your own database — checkout, renewals, seats and proration, refunds, claim-first webhook idempotency…

7 skillsRead →

agent-stack v0.13.2

agent orchestrators, their evals, the protocols they speak, and the wallet under LLM resale

Production patterns for AI agent systems, in four skills. The orchestrator: tool-calling loops that survive context pressure, pipelines with human checkpoints and resume, provider routing with fallback and…

4 skillsRead →

Install

One launcher installs every pack into every agent you have, and prunes the plain Claude Code copies that would shadow your plugins.

from npm — nothing to clone
$
npx sshlg-skills install
installed but behind
$
npx sshlg-skills update
what the current release of each member is
$
npx --yes sshlg-skills@latest list

Three commands are the whole interface. install when nothing is there, update when it is there and behind, list to see where each member stands.

update takes no member argument on purpose: a member updated on its own leaves the bundle in a combination nobody tested.

What it actually does

  • Claude Code → each pack as a plugin, never as a plain ~/.claude/skills/ copy.
  • Every other agent → the vercel skills CLI, into ~/.agents/skills/.
  • Then it prunes the plain Claude copies the skills CLI recreates on its own. That duplicate shadows your plugin and silently serves a stale skill — the one failure mode worth automating away.

Why a router, and not eleven more prompts

Installing a skill does not make an agent reach for it. The family writes a routing block into your agent's own instruction file: one table saying which pack answers what, and when. Every route also names the phrase that declines it, so the rule can be turned off in the sentence rather than in a file.

RouterAnswersWhen
super-uxwhat the interface must dothere is user-facing behaviour
sheleg-designhow it looks and movesthere is a visual layer
copywritinghow it soundstext a user of the product will read
sheleg-devwhat it runs on to charge, track and sign inmoney, tracking, sign-in or speed is being wired
agent-stackhow an agent system is built, judged and meteredthe thing being built is an agent
seo-llmowhether a machine will find ita logged-out reader can see the surface
evidence-docswhat proves itsomething is stated as true
task-pipelinehow the change reaches the repositorythe change touches the repository
project-auditwhat is actually true of this project right nowthe question is the whole project, not one change
make-skillhow the skill itself is builta skill or plugin changes shape
agent-syncwho is holding this filethe project has agent coordination on

Read every routing rule and its boundary →


The manifesto this pack implements

These skills are the reference implementation of Proof of Done: The Agentic Software Development Manifesto by Sergey Sheleg. Its claim is that the unit of progress is not generated code but an evidence-carrying change: one that carries the intent it implements, the evidence that verifies it, the limits of that evidence, and the decision that accepts it. The manifesto lives in its own repository; this one is where it stops being an argument.