maxgfr
UserAudit HTML/CSS/JSX against WCAG 2.2 AA and produce a dated compliance report, or author/review accessible markup. Zero-dependency static engine + agent judgment, with anti-hallucination check/verify gates.
Categories
Indexed Skills (5)
review-a11y
Use to REVIEW code changes for accessibility — '/review-a11y', 'review my changes/PR/branch for a11y', 'accessibility review of this diff'. Scopes to exactly the code under change (staged files, working diff, or branch vs merge-base), runs the bundled install-free ultra11y engine (node scripts/ultra11y.mjs, no keys; static WCAG 2.2 AA checks tied to success criteria, cross-file JSX/TSX via --graph), then adds YOUR judgment: refute false positives (SFC/library-source findings are preliminary), decide alt relevance and link purpose, name the rendering criteria (contrast, focus, zoom) as residual risks — never invent a non-conformity. Output: a severity-ranked review (blocking → major → minor) with file:line, concrete fixes, residual risks and a one-line verdict, in the conversation's language (technical tokens stay in English). Companion of the ultra11y skill (full audits, reports, PRD, RGAA packs). Triggers: 'review a11y', 'a11y review', 'accessibility review'.
ultra11y
Use to AUDIT existing HTML/CSS/JSX against WCAG 2.2 AA accessibility and produce a dated auditor-conformance report, OR to AUTHOR/REVIEW accessible markup (native-HTML-first, ARIA last). An install-free engine (`node scripts/ultra11y.mjs`, no keys) runs 72 static checks tied to WCAG criteria — alt/lang/title, unlabeled fields, empty links/buttons, tables, headings, ARIA vocabulary, label-in-name, autocomplete — measured against the official W3C ACT corpus. The engine decides 3 of the 55 criteria; the AI agent adjudicates the 38 judgment ones from harvested evidence + a per-criterion decision protocol (`verify --manual`, gated, fannable via `orchestrate`), and the 14 rendering ones go to `scan` — never silently conforming. Library/SFC code is audited as RENDERED captures (`render --setup`); country standards are pluggable packs (`--standard rgaa`, `scan --sample`). check/verify reject invented non-conformities. Triggers: 'audit WCAG/a11y', 'make accessible', 'fix a11y', 'audit RGAA'.
ultraindex
Use when an AI agent must UNDERSTAND a big repo, not merely search it — and its claims must be provable. ultraindex is the verified knowledge layer over codeindex, the deterministic zero-dep engine it vendors: codeindex tells you WHERE things are, ultraindex tells you what they MEAN and proves it. It builds an encyclopedia (a small always-loadable INDEX.md plus per-module entries): generated regions rebuild every time, human regions hold YOUR cited analysis, preserved across rebuilds and renames. dossier/ask hand you real source to write from; check REJECTS any [file:line] citation that does not resolve; verify proves each cited excerpt supports its claim. status is the enrichment work-queue, orchestrate fans it out to subagents. Triggers: index/map/document/analyze this codebase, where is X handled, how does Z work in this repo, which files do I change for Z, review this branch/PR, this repo is too big for context. Want plain code search, no model in the loop? Use codeindex directly.
reconstruct
Use when the user wants to rebuild, recreate, clone, or reverse-engineer an existing repository from scratch, or turn a codebase into specs/PRDs — e.g. "rebuild this project", "reverse engineer this repo", "generate a PRD/spec from this code", "recreate this app". ALSO use for greenfield asks — "build a new project from scratch", "turn my idea into PRDs / a build plan", "design a new app", "greenfield" — where there is no code yet and the facts are elicited through an interview. Works on any stack (JS/TS, Python, Ruby, Go, PHP, Java, mobile…). ALSO brainstorm product directions before building — "brainstorm ideas", "explore concepts", "compare approaches", "what should we build" — a divergent phase that generates several concepts and converges on one. Keywords: reconstruct, rebuild, clone, reverse engineer, from scratch, greenfield, build plan, new project, idea to PRD, brainstorm, ideation, explore concepts, compare approaches.
construct
Use when the user wants to turn a product idea into a serious, buildable requirements document (an SRD/PRD) — or build the app from one. Triggers: write an SRD or PRD, spec out a product, write/define requirements, idea to spec, brainstorm an idea, build from spec, one PRD per module, PRD folder. construct interviews the user, grounds every major decision in real research — competitors and market signal, comparable open-source projects and their issues/PRs, candidate-tech docs and StackOverflow pitfalls — then renders a complete SRD suite: vision, scope, functional requirements with Given/When/Then acceptance criteria, NFRs, data model, interfaces, ADRs, competitive landscape, build plan, traceability. Modules mode renders one PRD per module (prd/<module>/PRD.md); render --prd emits one PRD per requirement. A hard structural gate plus an advisory grounding report validate it; for building, it emits a BUILD-PLAN.json task DAG and construct verify referees the app against the SRD.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.