M4NUSH7

scrapling

Scrape, crawl, or extract structured data from websites with Scrapling - anti-bot bypass (Cloudflare Turnstile), stealth headless browsing, JavaScript rendering, adaptive selectors that survive page changes, and a Scrapy-like spiders framework. Use when a fetch is blocked, returns empty or JavaScript-rendered content, or hits an anti-bot wall; when data must be pulled from many pages or a whole site; or when asked to write Python scraping or spider code. NOT for simply reading, summarizing, or researching a public page that fetches fine - agent-reach owns that. Windows/CLI edition; the Linux sibling is scrapling-cowork.

ponytail

Forces the laziest solution that actually works - simplest, shortest, most minimal: YAGNI, stdlib before custom code, native features before dependencies. Intensity: lite, full (default), ultra. Use on ANY coding task - writing, refactoring, fixing, reviewing, or designing code, and choosing libraries. Also when the user says "ponytail", "be lazy", "lazy mode", "simplest/minimal solution", "yagni", "do less", "shortest path", or complains about over-engineering, bloat, or boilerplate. Subcommands: /ponytail (level), /ponytail-review (diff), /ponytail-audit (repo), /ponytail-debt (deferrals), /ponytail-gain (scoreboard), /ponytail-help. Not for non-coding requests: general knowledge, prose, translation, summaries, recipes.

agent-reach

Use when the user wants to research a topic, find sources, or search a platform by name - web pages, YouTube, GitHub, RSS/Atom feeds, V2EX, or Exa semantic search. Also when the user shares a URL from one of those platforms and wants it read, summarized, or checked for related content. Routes each request to the right CLI or API (curl plus Jina Reader, yt-dlp, gh, feedparser, mcporter plus Exa, V2EX) so results come back as raw source material, not narrowed by one search engine. Pairs with deep-research: use agent-reach to gather from several platforms, then hand the material over for synthesis. NOT for: writing the final report, analysis, or translation (it only fetches and searches); posting, commenting, or any other write action; Bilibili, which is not supported at all; Twitter/X, Reddit, Facebook, Instagram, LinkedIn, XiaoHongShu, or Xueqiu unless cookies or a login session are already set up - confirm setup first.

graphify

Use for tracing how code connects across files - "how does A reach B", "is there a route from X to Y", "what sits between these two modules", "what breaks if I change X" - and for orienting in an unfamiliar or large repo. Builds a queryable call/import graph once (graphify update), then answers with graphify path, affected, explain and query against that graph. Do NOT use when one bounded grep or one Read already answers the question, or on a repo small enough to read into context. Maps CODE structure, not agent workflows.

init-harness

One-shot initialization of a parallel-terminal Claude Code build harness in a project - a git worktree and branch per terminal, phase checkboxes as the task queue, pinned model/effort per agent, mechanical git-tag gates, hardened read/gate/log hooks, and a SQLite agent-memory shard per terminal. Use when the user says "/init-harness", "init-harness", "initialize the harness", "set up the build harness", "scaffold this project's harness", "bootstrap the agent harness", or drops in a project-context.yml and asks to initialize. Also use when a project needs multi-terminal agent orchestration with gates, or when an existing harness must be verified or repaired (hooks not firing, guards not blocking, .agents drift). Do NOT use for adding a single agent or skill to an existing project.

token-efficiency

Token optimization for Claude Code CLI sessions: RTK (Rust Token Killer) command-output compression, efficient file and command strategies, role-based model routing for subagents, Source-of-Truth keyword framework, hybrid SQLite agent memory with checkpoints, and concise output. Use when installing or verifying RTK or its hook, when a dev command's output is large (test runs, builds, installs, long logs), when spawning coding subagents, when writing agent logs or querying agent memory, when seeding a project's SoT keyword index, or when folding agent memory at a checkpoint. Also use when the user mentions 'rtk', 'token efficiency', 'token savings', 'context bloat', 'model routing', 'SoT keywords', or 'checkpoint agent memory', or asks for shorter or less verbose output, or to stop narrating or dumping whole files.

Web & Frontend Listed

ui-standout

Pick STANDOUT, fit-for-purpose UI components per use-case instead of reusing the same mundane flows. A 3-layer method: query real component catalogs (astryx / shadcn / Aceternity) for candidates, reject AI-slop defaults with a 46-rule taste gate (harvested from impeccable), and justify the pick with design principles (hierarchy, spacing, type, color, Von Restorff standout). Ships a use-case to component map so a checkout, dashboard, hero, onboarding or empty-state screen gets a standout component, not a stock grid. Use this BEFORE building, restyling, theming or polishing the rendered UI of a web, SSR or static app; when the user says a UI is bland/generic/mundane/"looks like every other app"; or when reviewing a built UI for slop. Not for backend/serverless-API/networking services (no UI), not for form validation/schema/handler code or tests that merely name a component, and not for charts specifically (the dataviz skill owns those).

Testing & QA Listed

playwright-tester

Author persistent Playwright .spec.ts tests, run them natively with `npx playwright test` (zero AI tokens per run), and fix failures in a diagnose-first loop. Use whenever the user wants to test a web app, check whether a frontend works, run or re-run UI tests, verify a form or login flow, do end-to-end testing, reproduce a UI bug, smoke test a site, catch a visual regression, check for broken links, test across browsers, add regression coverage, audit every component, sweep a codebase for UI bugs, or confirm recent changes did not break anything. Also triggers on "run my e2e tests", "my playwright tests are failing", "check the app in a browser", "does this page still work after my change", "test the happy path", "write e2e tests", "set up playwright", "find UI bugs", or "test the whole app and tell me what is broken". Prefer this over driving a browser live or screenshot-based checking for anything that should be verifiable more than once.

install-skill

End-to-end skill package installer and builder for Claude. Handles the FULL process: platform targeting (CLI / CLI+Desktop / Desktop), acquiring complete packages from GitHub repos or local folders (never just SKILL.md - all references, scripts, hooks, binaries, templates), relevance selection, personalization, platform adaptation, verification, packaging (.skill zips, staged folders), and install commands. Use this skill whenever the user wants to install a skill, install a skill package, add a skill from a repo or URL, build a skill package, port a skill between CLI and desktop, combine skills, or repackage/update an installed skill. Also use when the user complains a previous install only grabbed SKILL.md.

skill-creator

Author, audit and iterate on Claude skills - write a new SKILL.md from scratch, restructure or trim an existing one, and measure whether it actually works. Use when the user wants to create a skill, turn a workflow they just did into a skill, edit or improve an existing skill, fix a skill's frontmatter or description, run evals against a skill, benchmark a skill against a no-skill baseline, or optimize a description so it fires at the right time and not the wrong one. Also use when a skill needs validating before it ships - frontmatter key allowlist, the 1024-character description cap, ASCII, referenced-path existence. Not for acquiring or installing a skill from a repo or URL, or porting an already-authored skill between platforms - that is install-skill.