soumit-kaz
UserLazySitter - an autonomous AI engineering team: a 26-agent coding pipeline for Claude Code & Codex that turns a feature request into reviewed, tested, merge-ready code.
Categories
Indexed Skills (31)
lazysitter
Run the Autonomous Engineering Team (LazySitter) pipeline end-to-end for a feature request. Use this whenever the user asks to run LazySitter, run the autonomous engineering team, or drive a feature through the full intake→design→build→verify→release pipeline. You become the Tier-0 orchestrator — the only role with global visibility and the only one that spawns other agents.
a11y-audit
Audit a UI for accessibility against WCAG 2.2 AA — run the engine, then do the two-thirds it cannot see. Load when designing, building, or verifying any user-facing surface.
agent-supervision
Watch in-flight agents and intervene when one drifts — intent contracts, the intervention inbox, quarantine, and an honest account of which mechanisms actually enforce. Load when running a parallel wave, or when an agent looks like it is going the wrong way.
animation-motion
Add motion that is smooth, accessible and purposeful — compositor-friendly properties, reduced-motion, interruptible transitions, and what to animate at all. Load when a feature adds transitions, animations, or gesture-driven movement.
bundle-budget
Set and enforce a JavaScript bundle budget — measure the delta, attribute what grew, split at the right seams, and judge whether a dependency is worth its weight. Load when adding a dependency or reviewing bundle impact.
client-security
Client-side security for React apps — XSS sinks, secrets in the bundle, token storage, CSP, redirects, postMessage, and the limits of client-side enforcement. Load when handling untrusted data, credentials, or third-party code.
component-api-design
Design a component's public API — composition vs configuration, slots, polymorphism, controlled/uncontrolled, variant unions, escape hatches. Load when creating a reusable component or extending one others consume.
component-precedent
Decide reuse-vs-create for a component with evidence — build a ranked precedent set from the index, detect the mid-migration case, and cite by rank. Load before creating any component, and when auditing whether a created one should have existed.
core-web-vitals
Diagnose and protect LCP, INP and CLS — what each measures, what actually causes regressions, and how to verify. Load when a change affects a route's loading, interactivity, or layout.
data-fetching-cache
Design and audit client data fetching — query keys, invalidation edges, optimistic updates, out-of-order responses, retry and error surfacing. Load when a feature reads or mutates server data.
design-tokens
Work within the design system actually in force — find the token source, use the scales, detect violations, and handle theming. Load before writing any styles or choosing any visual value.
error-boundaries
Decide where errors are caught and what the user sees — boundary placement, what boundaries do and do not catch, recovery, and Suspense interaction. Load when adding a route, a data-fetching surface, or any component that can throw.
fe-index-query
Drive the LazySitter frontend index — the structural map of every component, hook, util, prop, call site and import edge in the repo. Use this INSTEAD of grep whenever the question is "does this exist", "who uses this", "what props does it take", "what would break if I change it", or "is this a duplicate". Load before any frontend exploration, review, or reuse decision.
forms-validation
Build forms that are accessible, resilient and honest about errors — labelling, validation timing, error association and announcement, submission states, and not losing the user's work. Load for any form.
hook-rules-audit
Audit React hook correctness — hook order, dependency arrays, stale closures, effects that should not exist, and cleanup. These are correctness bugs, not lint preferences. Load when writing or reviewing any component or custom hook.
hydration-safety
Find and fix hydration mismatches in server-rendered React — non-deterministic render values, browser-only APIs, locale and timezone divergence, invalid HTML nesting. Load for any SSR/Next work, and when debugging "works in dev, breaks in prod".
i18n-rtl
Build UI that survives translation and bidirectional text — no concatenated strings, no hardcoded formats, logical CSS properties, and layouts that tolerate 40% longer words. Load when the app is or may become multilingual.
image-media
Ship images and media correctly — dimensions, formats, responsive sources, lazy-loading, priority, and accessible alternatives. Load when a feature renders images, video, icons, or embeds.
keyboard-interaction
Implement the WAI-ARIA Authoring Practices keyboard contract for the widget you are building — dialog, menu, listbox, combobox, tabs, disclosure, tree, grid. Load before building any custom interactive widget.
loop-engineering
Run a bounded, observable loop — discovery, consensus, auto-fix, or a measured remediation loop — with an honest terminator and a structured round record. Load before starting any repeated round of work.
memory-leaks
Find and fix frontend memory leaks — missing teardown, growing caches, detached DOM, and closures holding trees alive. Load when a component subscribes to anything, and when a long-lived SPA degrades over time.
prop-analyzer
Analyse a component's prop contract against how it is actually used — dead props, undeclared props, boolean traps, required-vs-optional balance, spread leakage, and breaking-change cost. Load before designing, reviewing, or changing any component's public API.
render-performance
Diagnose and fix React re-render cost — identity churn, context fan-out, memoization that does and does not pay, and list reconciliation. Measure before optimizing. Load when a UI is slow, or when reviewing a change to a hot component.
responsive-layout
Make layouts work across widths, zoom levels, input types and content lengths — breakpoints, container queries, intrinsic sizing, space reservation, touch targets. Load when building or reviewing any layout.
rsc-boundary
Place and audit the Next.js App Router server/client boundary — 'use client' placement, serializable props, server-only modules, caching, streaming. Load for any Next App Router work.
state-topology
Decide where each piece of state lives — server, URL, client, session — and detect the misclassifications that cause most "wrong data on screen" bugs. Load before designing state for a feature, or when auditing an existing state layer.
test-selectors
Write frontend tests that assert user-visible behaviour and survive refactors — query priority, why role queries double as accessibility assertions, async waiting, and what never to assert. Load when writing or reviewing component tests.
ui-state-matrix
Enumerate and specify every state a UI surface can be in — loading, empty, error, partial, offline, permission, long content, concurrent. Load when writing a spec, and when verifying that each state was actually rendered.
utils-precedent
Find existing hooks, utils, formatters and validators — including renamed copy-paste clones no name search can find — and record the repo's real conventions for dates, numbers, errors and null handling. Load before creating any hook or utility.
virtualization
Decide whether a long list needs virtualization, and implement it without breaking accessibility, search, or scroll restoration. Load when rendering a list whose length is not bounded by design.
visual-regression
Catch unintended visual change — what to snapshot, how to make snapshots deterministic, how to review a diff, and why the highest-value finding is on a surface the feature never touched. Load when a render harness exists and a change touches shared UI.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.