hyperb1iss/hyperskills
7 skills4.2k installs91 starsGitHub
Install
npx skills add https://github.com/hyperb1iss/hyperskillsSkills in this repo
1Tui DesignTUI App Design Patterns Gallery is a research-style agent skill from hyperskills that catalogs how exceptional terminal applications solve layout, focus, and workflow problems. It is not a code generator; it gives solo builders and small teams concrete precedents—lazygit’s persistent multi-panel gold standard, lazydocker’s live dashboard split, contextual footers that eliminate memorized shortcuts, and overlay dialogs that preserve spatial context. Invoke it when you are sketching or refactoring a CLI tool, devops panel, or agent control surface and need defensible UX choices instead of one-off ncurses experiments. The material targets Go gocui-style architectures but patterns transfer to Bubble Tea, Ratatui, and other stacks. Pair reading this skill with your implementation framework docs so the agent proposes panel topology, keybinding discovery, and guided multi-step flows that users can navigate without a manual.788installs2OrchestrateOrchestrate is a journey-wide meta skill for solo and indie builders who have outgrown one-agent-at-a-time coding. It documents how to match orchestration strategy to work shape—whether you are running a research swarm, shipping independent features in parallel, chaining dependent tasks, applying the same refactor across repo partitions, auditing a codebase, or bootstrapping a greenfield project. The skill emphasizes partitioning by true independence, injecting enough shared context that parallel agents do not contradict each other, and choosing background versus foreground execution based on the dependency graph. Prompt structure is tied to agent role, and review-heavy pipelines are treated as adaptive overhead rather than a fixed tax. Activate it when you mention swarms, wave dispatch, or unleashing multiple agents, and use it whenever scale—not single-file edits—is the bottleneck across Validate prototyping spikes, Build implementation, Ship review, or Operate iteration.602installs3ResearchResearch is a journey-wide agent skill for solo and indie builders who need defensible answers before they bet time on a direction. It encodes a multi-agent, wave-based investigation model: prime the question, run a broad sweep, analyze gaps, deploy targeted follow-up waves, then synthesize only after evidence accumulates. That deferred synthesis matters because early conclusions from a shallow search often miss the nuance a later wave would surface. The framework is explicitly a shape to calibrate—not a ritual you must run at maximum depth every time. Quick mode trims the waves for small decisions; standard research uses one wave plus targeted follow-ups; deep dives use the full sequence. Triggers align with real builder language: investigate options, what is best, state of the art, deep dive, explore the landscape, and compare alternatives. Use it whenever a decision needs multiple sources—competitive positioning, library choice, integration patterns, or digging through an unfamiliar codebase—not only at the start of a project.590installs4BrainstormBrainstorm is a journey-wide agent skill for structured collaborative ideation using the Double Diamond model, with persistent memory so you do not endlessly re-explore settled decisions. It is meant to run before creative work—new features, greenfield projects, architecture choices, or design exploration—when mentions like brainstorm, ideate, design session, or “what should we build” appear. The skill teaches a flexible rhythm: ground in context, diverge on the problem, converge on a crisp problem statement, diverge on solutions, then converge on a plan, while skipping or revisiting phases as judgment dictates. A core thesis from 100+ real sessions is that agents excel at breadth and cross-domain connections while humans excel at selection; separating those modes produces better outcomes than one-shot “just build it” chat. Sibyl-backed memory reduces duplicate exploration. It is not a coding skill; it is a decision-quality skill you can invoke whenever commitment is expensive.565installs5Codex ReviewCodex Review Prompt Templates is an agent skill that packages battle-tested review prompts for solo and indie builders who want a second pair of eyes on a branch without wiring a full review bot. You pass the prompt text as the final argument to Codex, Claude in print mode, or any similar reviewer CLI, optionally piping the diff from `git diff main...HEAD` so the model focuses on what changed. The General Review pass is designed as a first sweep: it demands extreme thoroughness on logic and edge cases, injection and auth gaps, algorithmic and N+1 risks, and coupling or API consistency, while explicitly deprioritizing formatting and naming debates. Findings must cite exact locations, explain real bugs or risks, suggest fixes, and score confidence before an overall verdict of patch correct or incorrect. Additional templates such as the Security Deep-Dive let you stack focused passes after the broad one. It complements—not replaces—platform-native review commands when those already encode your team’s format.563installs6PlanPlan is a structured-planning agent skill for solo and indie builders who need more than bullet lists before touching code. It walks you through scoping, light exploration, decomposition, verification with approval, and ongoing tracking—optimized for agents that otherwise forget constraints mid-session. The methodology is mined from 200+ real planning sessions and stresses one core idea: abstract tasks fail; steps you can verify win. Sibyl integration is first-class so the plan does not die when the chat context rolls over. Use it when a feature feels large, cross-cutting, or politically fuzzy—not for one-line fixes you already know how to ship. Treat phases as rhythm, not bureaucracy: skip planning for trivial work, tighten loops when answers are obvious, and expect to replan as you learn. Pairs naturally with requirements clarity upstream and implementation skills downstream once VERIFY & APPROVE passes.559installs7ImplementImplement packages implementation benchmarks and archetypes distilled from tens of thousands of real agent-assisted coding operations. Instead of generic “move fast” advice, it cites measured patterns: how many reads and searches precede a change, how often to run typecheck versus lint versus tests, typical changes per commit, and where verification failures cluster. Debugging sections contrast quick one-iteration fixes with long spirals that often correlate with context overflow. Solo builders shipping with Claude Code, Cursor, or Codex can use it to calibrate session structure—explore before editing, verify every two or three edits, and avoid endless fix loops. It is meta guidance for execution quality, not a code generator; pair it with planning and review skills when you need specs or PR discipline.545installs