vosslab-skills

Detect and fix color contrast problems in source files and images: compute WCAG contrast ratios, then apply hue-preserving replacements to the failing hex values directly in the repo's source files so every color passes. Fixing the source is the primary outcome; the audit documents the result. Owns docs/PALETTE_CONTRAST_AUDIT.md end to end: writes or refreshes the per-repo palette audit from evidence gathered in the target repo this run, while citing the propagated generic docs/COLOR_CONTRAST_ACCESSIBILITY.md method doc. Use for color contrast, WCAG, accessibility, luminance, contrast ratio, palette audit, a11y, and dark mode contrast tasks: auditing and fixing a repo's color palette, applying accessible color replacements across source files, checking a specific foreground/background pair, darkening or brightening a failing color, spot-checking contrast in a screenshot or rendered image, generating a new set of accessible colors, or writing up a repo's palette contrast audit from real evidence.

Design, implement, verify, and debug Apple Liquid Glass surfaces in SwiftUI on macOS 26+ and iOS 26: `.glassEffect`, `GlassEffectContainer`, `glassEffectID` morphing, `.interactive()`, tint, layer order, and backdrop sampling. Use when glass looks flat, gray, missing, or invisible, when a screenshot cannot prove the effect is live, when deciding which surfaces should be glass versus stable content, or when guaranteeing text contrast over glass (Reduce Transparency, Increase Contrast, vibrancy, scrims, WCAG ratios). Covers capture-path hazards (cacheDisplay / bitmapImageRepForCachingDisplay / ImageRenderer omit backdrop compositing; capture on screen with screencapture), differential proof via Reduce Transparency, expected appearance per backdrop, and a drop-in GlassSurface wrapper with accessibility fallbacks.

Watchdog for background subagents that catches a stuck, stalled, or hung agent without killing a healthy one. Use when one or more subagents or background agents are running and you want periodic hang detection. Trigger phrases: hang check, agent stuck, agent hung, stalled subagent, watchdog, no output, monitor running agents.

Resolve TypeScript errors, eliminate `any`, and design modular, strict TypeScript types including generics, conditional types, mapped types, template literal types, branded or opaque types, and deep inference. Use for type-inference problems, `infer` or `extends` questions, utility types such as `Partial`, `Record`, `ReturnType`, `Awaited`, `NoInfer`, `satisfies`, module-boundary type design, function overloads, declaration merging, strict-mode refactors, and production-grade type-safety reviews.

Create, edit, and lint WeBWorK PG/PGML questions following docs/webwork guidance, HTML whitelist constraints, and renderer-based lint checks. Use for tasks like authoring new PGML problems, adjusting randomization or grading, fixing PGML rendering issues, and running renderer API linting.

Capture screenshots of a running app and embed them into README.md and docs/ to make GitHub landing pages novice-friendly. Classifies the app as PySide6 GUI, Swift GUI, terminal/CLI, or web app, captures with the appropriate backend (easy-screenshot for local windows, Playwright for web), writes PNGs to docs/screenshots/, and inserts embed lines into README/docs. Rewrites the readme-docs managed screenshot block (begin/end sentinels) with real embeds, idempotently. Use after readme-docs runs, when screenshots are stale or absent, or when adding a new app kind. Writes docs/screenshots/ PNG files and edits README/docs embed lines; runs in Wave 2 of the docset-updater refresh, after readme-docs. App code is read-only.

Refresh the whole repo doc set in one pass by invoking the per-doc skills in dependency order (`arch-docs`, `setup-install-usage-docs`, `readme-docs`, `related-projects-docs`, `news-release-docs`, `screenshot-docs`, `agents-md-fixer`), then audit any remaining `docs/` files those skills do not own. Use when the user wants all docs brought current at once, or the doc set as a whole is missing, drifted, or unaudited.

Create or refresh `docs/RELEASE_HISTORY.md` and `docs/NEWS.md` from `docs/CHANGELOG.md`, authoring two differentiated docs (full versioned dated log vs short curated highlights) and emitting a `/tmp` notes-file body for `devel/make_release.py --notes-file`. Use when preparing a release or when the release docs are drifted, empty, or out of sync with the changelog. Does NOT touch `docs/CHANGELOG.md`, `README.md`, or the broader doc set (use `docset-updater` for the rest).

Standardize `README.md` to match repo conventions: brief purpose, quick start, links to `docs/`, and a screenshots placeholder for the screenshot-docs skill. Use when `README.md` has drifted or is missing key pointers. Does NOT touch any file under `docs/` (use `arch-docs`, `setup-install-usage-docs`, or `docset-updater` for those).

Create or refresh `docs/RELATED_PROJECTS.md` from repo evidence plus bounded web discovery, writing a sourced, confidence-tiered map of upstreams, dependencies, integrations, sibling repos, and alternatives. Use when the user asks to document, discover, or refresh related or sibling projects, prior art, or alternatives for a repo. Does NOT touch `README.md`, `docs/CODE_ARCHITECTURE.md`, `docs/FILE_STRUCTURE.md`, `docs/INSTALL.md`, `docs/USAGE.md`, or the broader doc set (use `arch-docs`, `readme-docs`, or `docset-updater` for those).

Create or refresh `docs/INSTALL.md` and `docs/USAGE.md` from repo evidence, writing real content when install or usage evidence supports it and reporting gaps when it does not. Use when these two docs are missing, too thin, or stale. Does NOT touch `README.md`, `docs/CODE_ARCHITECTURE.md`, `docs/FILE_STRUCTURE.md`, or the broader doc set (use `readme-docs`, `arch-docs`, or `docset-updater` for those).

Trim `AGENTS.md` aggressively to a small pointer file (prefer ~15 lines, hard cap ~50) that points into `docs/*.md` with bare paths instead of restating rules. Use when `AGENTS.md` is bloated, philosophical, or duplicates style content already in `docs/*.md`. Does NOT touch `README.md` or files under `docs/` (use `readme-docs`, `arch-docs`, `setup-install-usage-docs`, or `docset-updater` for those).

Create or refresh `docs/CODE_ARCHITECTURE.md` and `docs/FILE_STRUCTURE.md` from current repo evidence. Use when the user asks to document or refresh repository architecture or file layout. Does NOT touch `README.md`, `docs/INSTALL.md`, `docs/USAGE.md`, or the broader doc set (use `readme-docs`, `setup-install-usage-docs`, or `docset-updater` for those).

Create, edit, and validate biology-problems bptools Python question generators and supporting YAML content. Use when requests involve authoring question scripts, updating files under problems/*-problems, tuning randomization or anti-cheat behavior, or debugging BBQ/QTI output that depends on bptools.py and qti_package_maker.

Design, implement, refactor, and review PySide6 desktop applications with strong widget architecture, signal-slot design, and state flow. Use when building or fixing Qt for Python windows, dialogs, forms, navigation shells, model-view tables, theming, or validation in Python GUI code.

Design, build, debug, and review SolidJS applications and their full stack -- core reactivity (`createSignal`, `createMemo`, `createEffect`, `createResource`), stores, control-flow components (`<For>`, `<Show>`, `<Switch>`), Solid Router, SolidStart, and Solid Meta. Use whenever the user works with SolidJS, Solid Router, SolidStart, or Solid Meta: signals not updating, "don't destructure props", `<For>` vs `<Index>`, `createStore` nested state, `useParams` / `action` / `createAsync` routing, `"use server"` server functions and data mutations, document head and SEO, or migrating React-shaped code to Solid's run-once reactive model.

Review, improve, and engineer UI/UX quality in any framework. Covers visual hierarchy, interaction quality, accessibility, validation, error states, empty states, and design heuristics. Use when reviewing or improving the user experience of an interface, regardless of the underlying technology.

Design, implement, debug, and review computer vision systems in Python, including image processing, detection, segmentation, classification, tracking, OCR, camera pipelines, and dataset-driven evaluation. Use when working with OpenCV, PyTorch vision models, video/image analysis, model-selection tradeoffs, annotation strategy, failure analysis, or CV performance and robustness problems.

Design, implement, debug, and review computational geometry algorithms in any language, including convex hull, polygon triangulation, Delaunay triangulation, Voronoi diagrams, polygon boolean clipping and overlay, segment intersection and sweep line, proximity and nearest-neighbor queries, point location, spatial data structures (k-d tree, BVH, R-tree, quadtree), arrangements, mesh processing, and distance/intersection queries in 2D, 3D, and nD. Use when working on geometric robustness, degeneracy and exact predicates, orientation and in-circle tests, winding and point-in-polygon, kernel/library choice (CGAL, Shapely/GEOS, SciPy spatial, Boost.Geometry), or geometry correctness, performance, and failure analysis.

Use when building a TypeScript browser game from modular `src/*.ts` files with parallel subagents to reduce wall-clock time. Layered on the `starter-repo-template/templates/typescript/` scaffold; this skill owns only the game overlay: workstreams, batched smoke testing, web-platform gotchas, single-file export, and GitHub Pages deploy. Immediate live target is `run_web_server.sh`; GitHub Pages and portable HTML export are post-show release paths. Delegates type design to `typescript-engineer`, manager dispatch to `delegate-manager-to-subagents`, and lane decisions to `parallel-plan`.

Create forward-looking implementation plans from scratch for coding teams without writing code. Use when the user needs a new plan document, major rewrite, milestone restructuring, acceptance gates, migration strategy, risk handling, or rollout planning for future work; produces parallel-plan-ready milestones when work can be split across independent workstreams; do not use this skill for post-implementation audits of an existing plan.

Use only when the user has an approved plan AND wants the main agent to manage execution through subagents instead of editing files directly. Skip when the user wants the main agent to deliver work directly as a doer rather than dispatching.

In-flight nudge to split current work into independent tracks for parallel subagent dispatch; does not create new plans (use blueprint-plan-drafter for that).

Parallel multi-reviewer audit launched before merge or release; not for single-pass review.

Use when the user invokes /stay-busy, asks to keep a manager/orchestrator/subagents busy, complains that an agent is waiting too much, or when an active delegate-manager-to-subagents workflow is about to idle despite safe evidence-producing follow-on work. Generates parallel workstreams with blocked fallbacks, ask-only boundaries, and required artifacts. Stay busy by producing evidence, not by creating motion. This is manager/orchestrator anti-idle behavior, not general productivity.

Review a specific existing implementation plan and verify whether code and tests actually satisfy the plan's goals, sequencing, and gates. Use when the user asks for plan quality review, closure validation, implementation-to-plan conformance, risk gap analysis, or go/no-go readiness on an existing plan document.

Implement the manager-prioritized milestone of an existing plan and carry it to closure, including adjacent fixes needed to pass gates and close the milestone cleanly. Use when a manager/reviewer says to start the next milestone and expects end-to-end delivery with evidence.

Gas Town style multi-agent coordination with role-mapped task routing and convoy-based work decomposition. Use when the user explicitly requests Gas Town workflow, convoy-style task decomposition, theatrical role-mapped coordination, or this repo's specific agent role system. Do not trigger on generic multi-agent or parallel task requests.

Single-pass Python correctness, security, and style review on demand; not for multi-reviewer audits before merge (use audit-code-reviewer for that).

Use when tasks involve reading, creating, or reviewing PDF files where rendering and layout matter; prefer visual checks by rendering pages (Poppler) and use Python tools such as `reportlab`, `pdfplumber`, and `pypdf` for generation and extraction.

Read specified repo rule files (AGENTS.md, docs/REPO_STYLE.md, docs/PYTHON_STYLE.md, docs/PYTEST_STYLE.md, docs/CLAUDE_HOOK_USAGE_GUIDE.md, docs/CHANGELOG.md) and answer targeted repo-rule questions, with missing-file handling and explicit-path checks. For docs/CHANGELOG.md, report only the latest dated entry. Use when a prompt asks for repo instruction answers about code execution, pytest style, Claude hook usage, or recent changes.

Guide for authoring Agent Skills (SKILL.md). Covers the open standard format, required frontmatter, directory layout, progressive disclosure, description writing, and best practices. Use when creating a new skill, improving an existing skill, or learning how skills work.

Generate thorough Python 3 pytest unit tests across a repo by scanning Python files and functions, writing one test module per source file, following current repo Python and pytest style, and documenting behavior that is unsafe or unclear to test.