tw93/waza
8 skills51.9k installs45k starsGitHub
Install
npx skills add https://github.com/tw93/wazaSkills in this repo
1Hunthunt is a failure-pattern reference from the Waza skill set for when a bug repeats, the first patch does not hold, or the symptom feels like runtime state rather than a typo in the file you just edited. Solo builders shipping SaaS, CLIs, or agent tooling use it as a checklist library: stale verifier caches pointing at deleted worktrees, scheduler logs without queue rows, source changes that never reach generated bundles, and auth guards defeated by delayed callbacks or relaunch paths. Each section pairs observable signals with concrete checks—prove the path exists before clearing caches, trace request through enqueue and persistence, verify build watches on source-to-artifact rules, and walk guard creation through invalidation. The skill does not replace reproduction steps; it reframes hunt so you test system boundaries and add regression tests at enqueue or artifact edges, not only inside worker bodies. Load it whenever chat-debugging has stalled and you need a named pattern to match.6.6kinstalls2ThinkThink is an agent skill for solo and indie builders who need a structured design pass before touching the codebase. It listens for planning and judgment triggers—whether in English or Chinese phrasing around 出方案, plan this, how should I design this, or whether something is worth building—and converts rough intent into one recommended direction or an executable handoff plan. The workflow emphasizes durable context preflight, evidence from the current repo and docs, and opinions that name what would falsify them. It is intentionally not a debugger or patch assistant; small edits and bug fixes are out of scope. Use it when architecture, feature shape, or value questions still have open decisions. After approval, you can proceed with implementation skills or your normal agent loop without re-deciding the same tradeoffs.6.6kinstalls3CheckCheck is a Waza agent skill for solo and indie builders who want structured review before merge, release, or publish—not open-ended brainstorming or debugging. It reads the actual diff, surfaces problems, fixes what is safe to fix in-tree, and escalates the rest. Triggers align with bilingual review intents (看看代码, 合并前, release, push, project audit) and maintainer workflows like closing issues or release reactions on GitHub. The outcome contract ties every run to evidence from the current change set plus in-session verification. It explicitly excludes exploring ideas, root-cause debugging, and generic prose review, so you reach for it when you already have changes worth gating. Pair it with planning skills when the user wants an approved plan implemented after review.6.6kinstalls4DesignDesign is an agent skill from tw93/waza that keeps solo builders from shipping generic AI UI slop by enforcing production structure and real variation discipline. It applies when you are building sidebar-and-workspace products (Slack, Linear, Notion class): surfaces step through background color and shadow only, interactions get consistent press scaling, and you pick one radius language per button family before writing JSX or SwiftUI. When stakeholders ask for options, the skill forces at least three meaningfully different directions—density, typographic voice, color temperature, layout, motion, decoration—not three hex swaps. For multi-page or production interfaces, you can emit a compact DESIGN.md-style brief up front covering theme, palette roles, typography, spacing, motion, and component rules so the agent codes against a single contract. Use it whenever you want intentional visual identity without hiring a design lead for every sprint.6.6kinstalls5Writewrite from the waza catalog helps solo builders edit English prose so it reads human: varied, imperfect, and specific instead of template-shaped AI output. It is organized as smells and examples aligned with SKILL.md’s core stance—match the problem, do not run a blind find-and-replace, and let the author’s voice and genre win. Agents apply nine core rules spanning filler removal, structural clichés, active voice with real human subjects, concrete nouns over vague declaratives, direct address, rhythmic variety, and cutting pull-quote bait. The file explicitly warns that over-editing is failure and that lists are illustrative, not mandatory sequences. Use whenever you draft or revise outward-facing copy across the journey, with the primary catalog home on growth content where polish affects trust and SEO.6.5kinstalls6HealthHealth is a Waza agent skill that treats your pasted workspace bundle as data only and runs a disciplined health review of how Claude Code is configured. It ingests CLAUDE.md layers, rules/, skill frontmatter and inventory, hooks, MCP status, handoff and memory files, then applies SIMPLE, STANDARD, or COMPLEX tiers so shallow setups are not over-audited and mature repos get deeper splits between CLAUDE.md, rules/, and skills. Checks emphasize executable guidance: build and test commands present, verification and compact-instruction sections where tiers demand them, and skill descriptions that state when to use and when not to use without overlapping triggers. For solo builders curating dozens of skills, Health is the periodic lint pass that catches context bloat, conflicting global rules, and skills that will misfire at invocation time. Run it whenever you change skills or CLAUDE.md, not only at project start.6.5kinstalls7ReadRead is a journey-wide agent skill that documents how to turn any URL into agent-ready Markdown using a ordered proxy cascade and local fallbacks. Solo builders install it when Claude Code, Cursor, or Codex must ingest docs, blog posts, or competitor pages without manual copy-paste or brittle scraping scripts. The workflow starts with defuddle.md for YAML-frontmatter-clean output, falls back to r.jina.ai for broader coverage and image links, optionally uses an installed web search plugin reader for JS-heavy sites, then runs npx agent-fetch --json or defuddle parse -m locally. GitHub file links get a dedicated fast path through raw content or the GitHub CLI so navigation chrome does not pollute the context window. Because the method is reference-only, you keep control of network calls and choose when to escalate past free proxies. Use it whenever an agent task depends on faithful off-site text, from idea-phase research through operate-phase incident writeups.6.4kinstalls8LearnLearn is a Waza agent skill for solo builders who need more than a quick summary when facing an unfamiliar domain, a pile of URLs, or mixed source files. Trigger phrases include 学习一下, 深入研究, 整理成文章, deep dive, compile sources, and one-stop reference requests. It runs a six-phase research workflow: collect or accept supplied materials, translate and explain, surface contradictions honestly, and shape output into research notes, an outline, a publish-ready draft, or a canonical reference—whatever mode fits the ask. The skill deliberately supports the user’s thinking rather than replacing it, and it draws a hard boundary against trivial single-URL fetch-only jobs, which belong elsewhere. Evidence expectations include cited sources, digestion notes, outline rationale, and self-review against the requested deliverable. For indie founders, it bridges early Idea research, Validate scoping narratives, and Grow content production without treating every question as a chat-sized answer.5.9kinstalls